CN118606855A - Soil contamination detection method and device based on artificial intelligence - Google Patents

Abstract

The invention provides a soil pollution degree detection method and device based on artificial intelligence, and relates to the technical field of data processing. Wherein the method comprises the following steps: acquiring sample data of sample soil, and marking the sample data; performing data expansion on the marked sample data to obtain expansion data; performing data processing on the extended data to obtain training data; inputting training data into a classifier to obtain a final classification model; after target data corresponding to target soil are obtained, carrying out data processing on the target data to obtain target data with characteristics reduced; inputting the target data with the characteristics reduced into a final classification model to obtain the pollution degree corresponding to the target soil; the soil pollution classification model is trained to detect the soil pollution state by the training data obtained after labeling, processing and expanding based on the sample data, so that the soil pollution detection cost is reduced, the detection time is shortened, and the application range is expanded.

Description

Soil contamination detection method and device based on artificial intelligence

Technical Field

The present invention relates to the field of data processing technology, and in particular to a soil contamination detection method and device based on artificial intelligence.

Background Art

With the rapid development of industrialization and urbanization, soil pollution has become a major environmental problem worldwide. The accumulation of heavy metals, organic pollutants and other harmful chemicals in the soil poses a serious threat to human health and the ecosystem. Therefore, it is particularly important to develop a method that can quickly and accurately determine the degree of soil pollution. Traditional soil pollution detection methods mostly rely on chemical analysis. Although these methods are accurate, they are usually time-consuming, costly, and not suitable for large-scale or real-time monitoring. In addition, these traditional methods are complicated to operate and require professional technicians for sample processing and data analysis, which limits their application in real-time and wide-area environmental monitoring.

Summary of the invention

In view of this, the purpose of the present invention is to provide a soil contamination detection method and device based on artificial intelligence. By training data obtained by annotating, processing and expanding sample data, a soil pollution classification model is trained to detect soil pollution status, thereby reducing the cost of soil pollution detection, shortening the detection time, and expanding the scope of application.

In a first aspect, the present invention provides a soil contamination detection method based on artificial intelligence, comprising: obtaining sample data of sample soil, and labeling the sample data; wherein the labeled category characterizes the degree of contamination of the sample soil; performing data expansion on the labeled sample data through a generative adversarial network based on variational quantum coding to obtain expanded data; performing data processing on the expanded data to obtain training data; wherein the data processing includes: feature extraction processing and feature dimension reduction processing; inputting the training data into a classifier to obtain a final classification model; wherein the classifier is a probabilistic neural network classifier based on adaptive learning rate adjustment; after obtaining the target data corresponding to the target soil, performing data processing on the target data to obtain the target data after feature reduction; inputting the target data after feature reduction into the final classification model to obtain the degree of contamination corresponding to the target soil.

In some preferred embodiments of the present invention, the step of training a generative adversarial network based on variational quantum coding includes: constructing a first quantum latent space through variational quantum coding; a generator of the generative adversarial network generates data based on features collected in the first quantum latent space; a discriminator of the generative adversarial network distinguishes between data generated by the generator and real data; and iteratively optimizing parameters of the generator and parameters of the discriminator through adversarial training until the diversity of data generated by the generator meets a preset diversity condition.

In some preferred embodiments of the present invention, data is generated based on an incremental loss function constraint in adversarial training; wherein the incremental loss function is based on the loss function of the generator, a first adjustment coefficient of the penalty term in the iteration process, and a second adjustment coefficient constraint of the penalty term in the iteration process.

In some preferred embodiments of the present invention, the diversity of the data generated by the generator is evaluated based on the generated data, the average value of the generated data, the number of generated data, and the dimension of the generated data.

In some preferred embodiments of the present invention, the step of processing the expanded data to obtain training data includes: extracting features from the expanded data through a neural network model optimized based on a submodular function to obtain feature-extracted data; wherein the optimization process of the submodular function is based on strategy constraints; inputting the feature-extracted data into a feature dimensionality reduction model to obtain reduced dimensionality data, and determining the reduced dimensionality data as training data; wherein the feature dimensionality reduction model is determined based on an autoencoding neural network algorithm of potential quantum coding.

In some preferred embodiments of the present invention, the steps of training a neural network model based on submodular function optimization include: initializing the parameters of the neural network model; setting the strategy constraints in the optimization process; wherein the strategy constraint function of the strategy constraint is based on the number of constraints, the weight corresponding to each constraint and the calculation formula constraints of each constraint; iterating the training operation until a preset maximum number of iterations is reached; the training operation includes: extracting features of the input data based on the neural network model; wherein the objective function of the feature extraction is based on the loss function of the input data, the sparsity constraint term, the hyperparameter for adjusting the sparsity and the hyperparameter constraints of the influence of the strategy constraints; evaluating the validity of the data after feature extraction, and adjusting the strategy constraints and the parameters of the neural network model based on the evaluation results; wherein the evaluation function is based on the number of evaluation indicator points and the calculation formula constraints of the evaluation indicators.

In some preferred embodiments of the present invention, the step of adjusting the parameters of the neural network model based on the evaluation results includes: adjusting the parameters of the neural network model based on a dynamic feature-aware network adjustment mechanism; wherein the adjustment function is based on the current parameters of the neural network model, the gradient of the evaluation function relative to the current parameters, the network structure adjustment function, the learning rate and the dynamic adjustment factor constraints; the gradient of the evaluation function relative to the current parameters is based on the extracted eigenvalues, the target eigenvalues and the total number of features constraints; the network structure adjustment function is based on the hyperparameters for controlling the network adjustment amplitude, the hyperparameters for controlling the network sensitivity and the optimization direction constraints for the generator.

In some preferred embodiments of the present invention, the step of training the feature dimensionality reduction model includes: initializing the parameters of the feature dimensionality reduction model; iteratively performing the data dimensionality reduction operation until a preset maximum number of iterations is reached; the data dimensionality reduction operation includes: forward propagating the original data of the input feature dimensionality reduction model to the second quantum latent space through the encoder to obtain the quantum bit data; based on the decoder, the quantum bit data is returned to the original space to obtain reconstructed data; based on the reconstructed data and the original data, the parameters of the encoder and the parameters of the decoder are updated using an asynchronous strategy; wherein, the state of the quantum bit is adjusted based on the restrictive optimization strategy.

In some preferred embodiments of the present invention, the training data is input into the classifier, and the step of obtaining the final classification model includes: dividing the training data into a training set and a validation set; training the classifier based on the training set, and validating the trained classifier based on the validation set until the loss function corresponding to the training set reaches a preset threshold, and determining the trained classifier as the final classification model; wherein, during the training process, the learning rate of the classifier is adjusted based on an adaptive learning rate adjustment mechanism.

In a second aspect, the present invention provides a soil contamination detection device based on artificial intelligence, comprising: a training data acquisition module, used to acquire sample data of sample soil and label the sample data; wherein the labeled category represents the degree of contamination of the sample soil; a data expansion module, used to perform data expansion on the labeled sample data through a generative adversarial network based on variational quantum coding to obtain expanded data; a data processing module, used to perform data processing on the expanded data to obtain training data; wherein the data processing includes: feature extraction processing and feature dimension reduction processing; a data classification module, used to input the training data into a classifier to obtain a final classification model; wherein the classifier is a probabilistic neural network classifier based on adaptive learning rate adjustment; a target data acquisition module, used to obtain the target data corresponding to the target soil, and then perform data processing on the target data to obtain the target data after feature reduction; a data evaluation module, used to input the target data after feature reduction into the final classification model to obtain the degree of contamination corresponding to the target soil.

The present invention brings the following beneficial effects:

The present invention provides a soil contamination detection method and device based on artificial intelligence, the method comprising: obtaining sample data of sample soil, and marking the sample data; wherein the marked category represents the pollution degree of the sample soil; performing data expansion on the marked sample data through a generative adversarial network based on variational quantum coding to obtain expanded data; performing data processing on the expanded data to obtain training data; wherein the data processing comprises: feature extraction processing and feature dimension reduction processing; inputting the training data into a classifier to obtain a final classification model; wherein the classifier is a probabilistic neural network classifier based on adaptive learning rate adjustment; after obtaining target data corresponding to the target soil, performing data processing on the target data to obtain target data after feature reduction; inputting the target data after feature reduction into the final classification model to obtain the pollution degree corresponding to the target soil; and detecting the soil contamination state by training a soil contamination classification model based on the training data obtained after marking, processing and expanding the sample data, thereby reducing the cost of soil contamination detection, shortening the detection time, and expanding the scope of application.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

FIG1 is a flow chart of a soil contamination detection method based on artificial intelligence provided by an embodiment of the present invention;

FIG2 is a flowchart of a method for training a generative adversarial network based on variational quantum coding provided by an embodiment of the present invention;

FIG3 is a schematic diagram of the structure of a soil contamination detection device based on artificial intelligence provided by an embodiment of the present invention;

FIG. 4 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention.

Icons: 310 - training data acquisition module; 320 - data expansion module; 330 - data processing module; 340 - data classification module; 350 - target data acquisition module; 360 - data evaluation module; 400 - memory; 401 - processor; 402 - bus; 403 - communication interface.

DETAILED DESCRIPTION

In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not require further definition and explanation in the subsequent drawings.

In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inside", "outside", etc. indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, or the positions or positional relationships in which the inventive product is usually placed when in use. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific position, be constructed and operated in a specific position, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", "third", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

In addition, the terms "horizontal", "vertical", "overhanging" and the like do not mean that the components are required to be absolutely horizontal or overhanging, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", and does not mean that the structure must be completely horizontal, but can be slightly tilted.

In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

With the rapid development of industrialization and urbanization, soil pollution has become a major environmental problem worldwide. The accumulation of heavy metals, organic pollutants and other harmful chemicals in the soil poses a serious threat to human health and the ecosystem. Therefore, it is particularly important to develop a method that can quickly and accurately determine the degree of soil pollution. Traditional soil pollution detection methods mostly rely on chemical analysis. Although these methods are accurate, they are usually time-consuming, costly, and not suitable for large-scale or real-time monitoring. In addition, these traditional methods are complicated to operate and require professional technicians for sample processing and data analysis, which limits their application in real-time and wide-area environmental monitoring.

Some technical means are disclosed in the prior art. By dividing the acquisition path of the acquisition unit through the central control unit, the accuracy of soil samples can be guaranteed, the error in soil pollution monitoring can be reduced, and the impact on water pollution monitoring can be reduced. The heavy metal content of the water samples and riverbed samples detected by the detection unit is compared with the heavy metal content of the soil samples to confirm the source of soil pollution. The detection path and sample quantity of the soil samples can be intelligently calculated, and the source of soil pollution can be determined and the abnormality can be located in time. It can quickly provide an investigation and save a lot of time cost, realize the rapid division of the path, thereby improving the efficiency of soil detection, and further avoiding the soil pollution caused by water pollution and the resulting crop pollution.

However, in the existing technology, during the training process, the training samples are often limited by the authenticity and diversity of the data, making it difficult to effectively improve the generalization ability of the model.

Based on this, the present invention provides a soil contamination detection method and device based on artificial intelligence. Some embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. The following embodiments and features in the embodiments can be combined with each other without conflict.

Embodiment 1

The embodiment of the present invention provides a soil contamination detection method based on artificial intelligence. Referring to FIG. 1 , a flow chart of a soil contamination detection method based on artificial intelligence provided by the embodiment of the present invention is shown. The method includes:

Step S102, obtaining sample data of the sample soil, and labeling the sample data; wherein the labeled category represents the degree of contamination of the sample soil.

Specifically, it is necessary to collect soil samples and conduct laboratory analysis to obtain sample data. This data may include chemical composition, physical properties, biological indicators, etc. Subsequently, each sample is labeled with the degree of pollution, which can be done through expert knowledge or reference to existing environmental standards. The labeled results will serve as the basis for subsequent machine learning model training.

In some preferred embodiments of the present invention, soil pollution data is collected from soil samples in multiple geographical areas, covering different pollution levels, including industrial areas, agricultural areas and residential areas, and the sampling method adopts random sampling to ensure the representativeness and diversity of the samples.

In some preferred embodiments of the present invention, soil samples are collected at different locations using automated sampling equipment. These equipment can be equipped with a GPS positioning system to ensure accurate record of the sampling location; the soil samples are sent to a laboratory for detailed analysis, including determination of parameters such as organic pollutants, inorganic content, and heavy metal content; experts classify and label the degree of contamination of each sample according to current environmental standards and soil quality guidelines.

In some preferred embodiments of the present invention, the collected data attributes include: a1: pH value, a2: organic matter content (%), a3: lead content (ppm), a4: mercury content (ppm), a5: arsenic content (ppm), a6: cadmium content (ppm), a7: nitrogen content (%), a8: phosphorus content (%), a9: potassium content (%) and a10: total heavy metal content (ppm); it should be emphasized that this embodiment is only to illustrate a data format and type of the present invention. In practical applications, the attributes of the data are usually more than 10 attributes, and the number of attributes of the data may reach dozens or even hundreds. Further, the collected data is annotated, and the annotation method of the present invention is manual annotation. In some preferred embodiments of the present invention, the annotated categories include "unpolluted", "mildly polluted", "moderately polluted" and "severely polluted".

Step S104, performing data expansion on the labeled sample data through a generative adversarial network based on variational quantum coding to obtain expanded data.

Specifically, in order to improve the generalization ability of the model, a generative adversarial network based on variational quantum coding is used to expand the existing data. This technology can generate new, diverse and realistically distributed soil sample data, thereby increasing the scale and diversity of the dataset.

In some preferred embodiments of the present invention, concepts based on quantum computing are introduced, data is encoded by quantum bits (qubits), and the principles of quantum entanglement and superposition are used to improve the efficiency and capability of data processing.

Step S106, processing the expanded data to obtain training data; wherein the data processing includes: feature extraction processing and feature dimension reduction processing.

Specifically, data processing includes two parts: feature extraction and feature dimensionality reduction. Feature extraction refers to extracting information that is helpful for classification tasks from raw data, while feature dimensionality reduction reduces the complexity of data, removes redundant information, and improves computational efficiency.

In some preferred embodiments of the present invention, advanced data processing techniques such as principal component analysis (PCA), wavelet transform or deep learning feature extractor are applied to extract key information from the raw data.

Step S108, inputting the training data into the classifier to obtain a final classification model; wherein the classifier is a probabilistic neural network classifier based on adaptive learning rate adjustment.

Specifically, the processed data is input into a classifier for training. Here, a probabilistic neural network classifier based on adaptive learning rate adjustment is used. Adaptive learning rate can help the model converge to the optimal solution faster and improve training efficiency.

Step S110, after acquiring the target data corresponding to the target soil, the target data is processed to obtain the target data after feature reduction.

Specifically, when a new target soil sample needs to be tested for contamination, the target data must first be processed in the same way as the training data to ensure feature consistency.

Step S112, inputting the target data after feature reduction into the final classification model to obtain the pollution degree corresponding to the target soil.

Specifically, the processed target data is input into the trained final classification model, and the model will output the pollution degree of the target soil. This result can be used in many fields such as environmental monitoring and agricultural management.

In some preferred embodiments of the present invention, after the model training is completed, the newly collected soil pollution detection data samples are processed through feature extraction and feature dimension reduction steps, and the processed features are input into the classifier to classify the soil pollution. In one embodiment, the classification categories include "unpolluted", "mildly polluted", "moderately polluted" and "heavily polluted".

The present invention provides a soil contamination detection method based on artificial intelligence, comprising: obtaining sample data of sample soil, and marking the sample data; wherein the marked category represents the pollution degree of the sample soil; performing data expansion on the marked sample data through a generative adversarial network based on variational quantum coding to obtain expanded data; performing data processing on the expanded data to obtain training data; wherein the data processing includes: feature extraction processing and feature dimension reduction processing; inputting the training data into a classifier to obtain a final classification model; wherein the classifier is a probabilistic neural network classifier based on adaptive learning rate adjustment; after obtaining the target data corresponding to the target soil, performing data processing on the target data to obtain the target data after feature reduction; inputting the target data after feature reduction into the final classification model to obtain the pollution degree corresponding to the target soil; and training the soil pollution classification model to detect the soil pollution state by the training data obtained after marking, processing and expanding the sample data, thereby reducing the cost of soil pollution detection, shortening the detection time, and expanding the scope of application. Through artificial intelligence technology, a large number of soil samples can be analyzed in a relatively short time, and the prediction results of the model have high reliability. In addition, this method has good scalability and can adjust model parameters according to different application scenarios to meet different needs.

Embodiment 2

On the basis of the above embodiments, the embodiments of the present invention provide another soil contamination detection method based on artificial intelligence, focusing on describing the steps of data expansion, feature extraction, feature dimensionality reduction and classifier training.

It is understandable that the collection, acquisition, labeling and preprocessing of training data are time-consuming and labor-intensive, and insufficient training samples can easily lead to poor generalization ability of the model, while affecting the accuracy of the model. The embodiment of the present invention proposes a generative adversarial network algorithm based on variational quantum coding, which uses the advantages of quantum computing to enhance the ability of the generative model, and at the same time achieves more refined control of the generation effect through an incremental loss function. It not only considers the global difference between the generated data and the real data, but also guides the model to pay attention to the details of the generation process through gradually increasing penalty terms, so as to obtain more realistic data after multiple iterations, so as to improve the model's learning ability for small sample data and the diversity and authenticity of the generated data.

Specifically, in some preferred embodiments of the present invention, referring to FIG. 2 , a flowchart of training a generative adversarial network based on variational quantum coding provided by an embodiment of the present invention specifically includes:

Step S202, constructing a first quantum latent space by variable component quantum coding.

Specifically, first, a quantum latent space is constructed using variational quantum coding technology. This space is high-dimensional and allows the storage and processing of a large amount of quantum state information. Through quantum circuit design, soil sample data is mapped to quantum bits (qubits) to form quantum states. These quantum states contain complex relationships and patterns of the original data. Using quantum machine learning algorithms, such as variational quantum feature extraction, useful features are extracted from the quantum states and the first quantum latent space is constructed.

In some preferred embodiments of the present invention, in the initial stage, a latent space is constructed by a variational quantum algorithm, and the latent space has high-dimensional characteristics, providing a richer data representation basis for subsequent generation; specifically, it includes defining the construction process of the quantum latent space, which is implemented by variational quantum coding. Considering that the goal of variational quantum coding is to minimize the distance between the quantum state and the target state, the embodiment of the present invention uses the following method to represent this process:

The quantum state is represented as: |ψ(b _x )>, where b _x is the feature vector extracted from the data set.

The target state is expressed as: |φ(by ₎ >, where _by is the target feature vector, that is, the data preprocessed by some normalization.

The distance between the quantum state and the target state is measured by FidelityF, which is calculated as formula (1):

F＝|<ψ(b _x )|φ(b _y )>| ² (1)

Here, the goal is to maximize F by adjusting the parameters b _θ in the quantum encoding process.

Furthermore, the construction method of the quantum state |ψ(b _x )> can be expressed as formula (2):

|ψ(b _x )>＝U(θ)|0> ⁿ (2)

Wherein, U(θ) is a parameterized quantum circuit applied to the initial quantum bit state |0> ⁿ (n represents the number of quantum bits), and θ is a parameter of the circuit, which is adjusted by a classical optimization algorithm to minimize the objective function. In some preferred embodiments of the present invention, the number of quantum bits n is set to 5.

Moreover, the construction method of the target state |φ(b _y )> can be expressed as formula (3):

Among them, c _k is the coefficient calculated based on the target feature _by , K is the number of ground states, and |k> is the quantum ground state.

Furthermore, based on the constructed quantum latent space, the framework of generative adversarial network is used for training. The generator is responsible for sampling and generating data from the quantum latent space, while the discriminator attempts to distinguish the generated data from the real data. Through adversarial training, the parameters of the generator and discriminator are continuously optimized to improve the authenticity and diversity of the generated data.

Step S204: a generator of a generative adversarial network generates data based on features collected in the first quantum latent space.

Specifically, in the constructed first quantum latent space, potential features are collected through quantum sampling, which can represent the key information of soil samples. The generator of the generative adversarial network (GANs) receives these potential features as input and learns how to generate synthetic data similar to real soil sample data. The generator consists of a series of neural network layers that convert potential features into synthetic data with a distribution similar to real sample data.

In some preferred embodiments of the present invention, in the framework of a generative adversarial network based on variational quantum coding, the calculation method of the generator loss function _LG is expressed as formula (4):

L _G =-log D(G(b _z ))+λ(1-F) (4)

Wherein, b _z is a noise vector sampled from the quantum latent space, and λ is a balancing term used to adjust the weight between the adversarial loss and the quantum coding distance. In some preferred embodiments of the present invention, the balancing term λ is set to 0.1.

Step S206, the discriminator of the generative adversarial network distinguishes the difference between the data generated by the generator and the real data.

Specifically, the task of the discriminator of the generative adversarial network is to distinguish between the synthetic data generated by the generator and the real soil sample data. The discriminator is also a neural network that evaluates the authenticity of the generated data by learning to identify small differences in the data. The output of the discriminator is a probability value that indicates the possibility that the data is a real sample.

In some preferred embodiments of the present invention, in the framework of a generative adversarial network based on variational quantum coding, the calculation method of the discriminator loss function _LD is expressed as formula (5):

L _D =-[log D(b _y )+log (1-D(G(b _z )))] (5)

Furthermore, during adversarial training, an incremental loss function is used to refine the quality of generated data. The incremental loss function adds additional penalty terms in each iteration, so that the model pays more attention to the details and quality of generated data, thereby achieving the generation of high-quality data.

Step S208, iteratively optimize the parameters of the generator and the parameters of the discriminator through adversarial training until the diversity of the data generated by the generator meets the preset diversity conditions.

Specifically, during adversarial training, the generator and the discriminator undergo an iterative competition and optimization process. The generator tries to generate more and more realistic data to "cheat" the discriminator, while the discriminator strives to improve its ability to distinguish between true and false. Through repeated training iterations, the generator gradually learns to generate diverse and highly realistic soil sample data. Training continues until the generated data meets the predetermined diversity conditions, that is, the generated data is rich enough to cover the main features and distribution of real data.

In some preferred embodiments of the present invention, the incremental loss function is designed to gradually improve the detail quality of the generated data in each iteration, and the incremental loss function _Linc is defined as formula (6):

Wherein, _ΔLn is the penalty term added at the nth iteration, α is the first adjustment coefficient of the incremental loss function, _βn is the second adjustment coefficient of the incremental loss function, and N is the number of iterations. In some preferred embodiments of the present invention, α is set to 0.5, and the sequence of _βn increases exponentially, with the starting value set to 0.05.

Furthermore, the calculation method of _ΔLn can be expressed as formula (7):

in, To generate data relative to the input The gradient of is the gradient of the true data with respect to the same input.

Through the above steps, the generative adversarial network based on variational quantum coding can effectively expand the soil sample data set and provide more comprehensive and representative data for subsequent model training. This not only helps to improve the generalization ability of the model, but also can solve the problem of insufficient data caused by actual sampling difficulties or high costs to a certain extent.

Furthermore, in some preferred embodiments of the present invention, data is generated based on an incremental loss function constraint in adversarial training; wherein the incremental loss function is based on the loss function of the generator, a first adjustment coefficient of the penalty term in the iteration process, and a second adjustment coefficient constraint of the penalty term in the iteration process.

Specifically, after each round of training, feedback adjustment is performed based on the difference between the generated data and the real data. The embodiment of the present invention uses the following method to represent this process:

The difference measure D _diff between the generated data and the real data is calculated and can be expressed as formula (8):

D _diff ＝∥G(b _z )-b _y ∥ ₂ (8)

Adjust the quantum coding parameters b _θ and loss function weights α, β _n according to D _diff .

Furthermore, in some preferred embodiments of the present invention, the diversity of the data generated by the generator is evaluated based on the generated data, the average value of the generated data, the number of generated data, and the dimension of the generated data.

Specifically, in order to ensure the diversity of generated data, the present invention uses the diversity index D _var for evaluation, which is calculated as formula (9):

Among them, M is the number of generated data, d is the data dimension, is the mean vector of all generated data.

Furthermore, we generate the mean vector of the data The calculation method can be expressed as formula (10):

in, represents the i-th generated data vector, and M is the total number of generated data. By calculating the average value of all generated data vectors, we get the mean vector Used to evaluate the diversity of generated data. In some preferred embodiments of the present invention, M is set to 1000.

The soil contamination detection method based on artificial intelligence provided by the embodiment of this aspect performs diversity assessment on the generated data. If the data diversity is insufficient, the parameters of the quantum latent space are adjusted to increase the variation range of the generated data, thereby ensuring that the ultimately generated data is both real and diverse.

Furthermore, the prior art may not be able to effectively filter out irrelevant features in feature extraction, affecting the performance and accuracy of the model. Traditional feature dimensionality reduction technology may lose key information during data compression, affecting subsequent data processing and analysis. In some preferred embodiments of the present invention, the expanded data is processed to obtain training data, including the following steps A1 to A2:

Step A1, extracting features from the expanded data through a neural network model based on submodular function optimization to obtain feature-extracted data; wherein the optimization process of the submodular function is constrained based on a strategy.

Specifically, first, we need to design a neural network model based on submodular function optimization to extract useful features from the augmented data. Submodular function is a mathematical tool commonly used in optimization problems. It can ensure that the combination of solutions is still a solution when certain conditions are met; secondly, in the training process of the neural network, policy constraints are introduced. These constraints are based on submodular functions to ensure that the feature extraction process can capture the most meaningful information in the data while avoiding overfitting; thirdly, the augmented data is input into the designed neural network model, and features that can represent the important characteristics of the original data are extracted through forward propagation calculation.

In some preferred embodiments of the present invention, training the neural network model based on submodular function optimization includes the following steps B1 to B3:

Step B1, initializing the parameters of the neural network model.

Specifically, before starting training, you first need to initialize the parameters of the neural network, including weights and biases. These parameters are randomly generated, usually using a specific distribution, such as normal distribution or uniform distribution. The purpose of initialization is to provide a starting point for the network so that it can begin the learning process. A good initialization method can help the model converge faster and avoid some common problems, such as gradient disappearance or gradient explosion.

In some preferred embodiments of the present invention, the parameters of the neural network model are initialized, and Represents the initial parameter set of the neural network, where the subscript 0 represents the initial state before training begins.

Step B2, setting the strategy constraints in the optimization process; wherein the strategy constraint function of the strategy constraint is based on the number of constraints, the weight corresponding to each constraint and the calculation formula constraint of each constraint.

Specifically, during the optimization process, policy constraints are introduced to guide the learning process of the neural network. These constraints are based on submodular functions, and they define the solution space of the optimization problem. Each constraint has a corresponding weight, which determines the importance of the constraint in the optimization process. The higher the weight, the greater the impact of the corresponding constraint on the model. The calculation formulas of the constraints are pre-defined, and they can be about certain properties of the data, such as sparsity, smoothness, or other characteristics related to the problem.

In some preferred embodiments of the present invention, strategic constraints are set in the optimization process. These constraints are formulated based on the specific requirements of soil contamination detection to ensure that the extracted features can accurately reflect the soil contamination. Specifically, the strategic constraint C(c _θ ) is set, where c _θ is a network parameter. The constraint function is formalized as formula (11):

Where N _c is the number of constraints, c _i is the weight for the i-th constraint, and _fi (c _θ ) is the calculation formula for the i-th constraint, mapping the network parameters to the degree of satisfying a specific constraint.

In some preferred embodiments of the present invention, for each constraint condition _fi (c _θ ), a constraint is provided to maintain the extracted features to have high sensitivity for identifying excessive pH in soil contamination detection, and the constraint can be embodied as formula (12):

in, is the part of the network weight related to the constraint _ci , X is the input data, Y _crack is the expected output of the pH over-limit feature, and σ is the hyperparameter controlling the sensitivity.

Step B3, iteratively perform training operations until a preset maximum number of iterations is reached; the training operations include: extracting features from input data based on a neural network model; wherein the objective function of feature extraction is based on a loss function of the input data, a sparsity constraint term, a hyperparameter for adjusting sparsity, and a hyperparameter constraint for the influence of strategy constraints; evaluating the validity of the data after feature extraction, and adjusting strategy constraints and parameters of the neural network model based on the evaluation results; wherein the evaluation function is based on the number of evaluation indicator points and the calculation formula constraints of the evaluation indicators.

Specifically, before reaching the preset maximum number of iterations, the following training operations are performed continuously:

Feature extraction: Extract features from the input data using the neural network model with the current parameters. The goal of this step is to maximize the objective function, which is based on the loss function of the input data, the sparsity constraint term, the hyperparameters that adjust the sparsity, and the hyperparameters that constrain the influence of the policy constraints. Evaluate effectiveness: After extracting the features, the effectiveness of these features needs to be evaluated. This can be done by calculating the evaluation function, which is based on the number of evaluation indicator points and the calculation formula constraints of the evaluation indicators. Adjust parameters: Based on the evaluation results, adjust the policy constraints and the parameters of the neural network model. This is achieved through the back-propagation algorithm, which calculates the gradient of the loss function with respect to the model parameters and updates the parameters accordingly.

In some preferred embodiments of the present invention, a neural network is used to extract features from input data, and a sparsity constraint is introduced in the feature extraction process. The extracted features are continuously adjusted by iteratively optimizing the network parameters until a predetermined sparsity level is reached. Specifically, the feature extraction objective function is set as formula (13):

O(c _θ ,C)=L(c _θ )+λs·S(c _θ )+γs·C(c _θ ) (13)

Wherein, L(c _θ ) is a data-based loss function used to evaluate the effect of feature extraction; S(c _θ ) is a sparsity constraint term, λs is a hyperparameter for adjusting sparsity, and γs is a hyperparameter for adjusting the influence of strategy constraints. In some preferred embodiments of the present invention, λs and γs are set to 0.2 and 0.8 respectively.

In some preferred embodiments of the present invention, the calculation method of the sparsity constraint term S(c _θ ) can be expressed as formula (14):

Among them, ∥c _θ ∥ ₁ is the L1 norm of parameter c _θ , which is used to promote the sparsity of network parameters, thereby enhancing the interpretability of the model and reducing the risk of overfitting.

Through the above steps B1 to B3, the neural network model gradually learns and extracts useful features under the guidance of policy constraints. This method based on submodular function optimization enables the model to effectively learn meaningful feature representations from data while meeting specific constraints. This not only helps to improve the performance of the model, but also ensures the interpretability and generalization ability of the model.

Furthermore, after each round of training, the effectiveness of the extracted features for the soil contamination detection task is evaluated, and the strategy constraints and network parameters are adjusted according to the evaluation results to further optimize the feature extraction process. Specifically, the evaluation function is used to adjust the network parameters and strategy constraints, and the evaluation function is formula (15):

Among them, M is the number of evaluation indicators, _gj is the calculation formula of the jth evaluation indicator, which depends on the output of the feature extraction objective function.

In some preferred embodiments of the present invention, each evaluation index _gj can be embodied as a performance metric of the model on a specific task. For example, the accuracy evaluation index for the pH over-limit detection task can be formula (16):

Wherein, TP is the number of true positive examples, indicating the number of correctly identified features of excessive pH; FP is the number of false positive examples, indicating the number of features incorrectly identified as excessive pH. Wherein, the identification method is implemented by a preset Softmax function.

Furthermore, during the iteration process, the parameter update method can be expressed as formula (17):

Among them, ηs is the learning rate, is the gradient of the objective function with respect to the network parameters, and t represents the current iteration number. In some preferred embodiments of the present invention, the learning rate ηs is set to 0.01.

The embodiment of the present invention proposes a neural network algorithm based on strategy-constrained submodular function optimization, which uses strategy constraints to guide the submodular function optimization process to ensure that the feature extraction process is both efficient and can retain the core information of the data. By adding constraints in the optimization process, the most discriminative features for soil contamination detection can be effectively extracted, while avoiding the interference of irrelevant features. In addition, the present invention introduces sparsity constraints in the feature extraction process to reduce the dimension of the feature space and improve the interpretability of the model, which not only helps to reduce the computational complexity, but also can significantly improve the accuracy and efficiency of feature extraction when processing large-scale data.

Furthermore, in some preferred embodiments of the present invention, the step of adjusting the parameters of the neural network model based on the evaluation results includes: adjusting the parameters of the neural network model based on a dynamic feature-aware network adjustment mechanism; wherein the adjustment function is based on the current parameters of the neural network model, the gradient of the evaluation function relative to the current parameters, the network structure adjustment function, the learning rate and the dynamic adjustment factor constraints; the gradient of the evaluation function relative to the current parameters is based on the extracted eigenvalues, the target eigenvalues and the total number of features constraints; the network structure adjustment function is based on the hyperparameters for controlling the network adjustment amplitude, the hyperparameters for controlling the network sensitivity and the optimization direction constraints for the generator.

Specifically, in some preferred embodiments of the present invention, considering the implementation of the dynamic feature perception network adjustment mechanism, the feature extraction effect evaluation function of the model after the tth iteration is set to D _t (c _θ ,Y _target ), where Y _target is the target feature vector, and the evaluation function is defined as formula (18):

in, is the kth extracted eigenvalue, is the kth target feature value, and K is the total number of features.

Furthermore, based on the result of D _t (c _θ ,Y _target ), the dynamic feature-aware network adjustment mechanism adjusts the network parameters as shown in formula (19) by the following formula:

in, is the gradient of the evaluation function relative to the network parameters; Adjust(c _θ ,D _t ) is the network structure adjustment function based on the current iterative feature extraction effect, defined as formula (20):

Adjust(c _θ ,D _t )=α·exp(-β·D _t (c _θ ,Y _target ))·δc _θ (20)

Among them, α is a hyperparameter that controls the adjustment amplitude of the network, β is a hyperparameter that controls the sensitivity of the network; δc _θ is the optimization direction for D _t , which aims to reduce the difference between the feature extraction effect and the target; ζ is a dynamic adjustment factor used to balance the original gradient descent and the network structure adjustment based on the feature extraction effect.

During the training process, the model is adjusted through a dynamic feature perception network, that is, the network parameters are updated by dynamically evaluating the changes in features, thereby achieving a model that is more sensitive and adaptable to the soil contamination detection task. Specifically, during the evaluation and adjustment after each iteration, the network structure is optimized based on the feature extraction effect of the current iteration to ensure that the model can more accurately capture the features that are critical to the current soil contamination detection task.

Step A2, inputting the data after feature extraction into the feature dimension reduction model to obtain the reduced data, and determining the reduced data as training data; wherein the feature dimension reduction model is determined based on the autoencoder neural network algorithm of potential quantum coding.

Specifically, first, we need to build an autoencoder neural network algorithm based on potential quantum coding. The autoencoder is an unsupervised learning neural network that can learn an effective representation (encoding) of data and reconstruct data from this representation; secondly, the feature-extracted data is input into the autoencoder neural network, and the network compresses the data into a low-dimensional representation, that is, the reduced-dimensional data. This low-dimensional representation retains the most critical information of the original data while removing noise and redundancy; thirdly, the reduced-dimensional data is determined as training data for training the final classification model. These data are now more suitable for training because they contain both necessary information and reduce the computational burden.

Furthermore, in some preferred embodiments of the present invention, training the feature dimensionality reduction model includes steps C1 to C2:

Step C1, initializing the parameters of the feature dimensionality reduction model.

Step C2, iteratively performing the data dimension reduction operation until a preset maximum number of iterations is reached; the data reduction operation includes: forwarding the original data of the input feature dimension reduction model through the encoder to the second quantum latent space to obtain the quantum bit data; based on the decoder, the quantum bit data is converted to the original space to obtain reconstructed data; based on the reconstructed data and the original data, the parameters of the encoder and the parameters of the decoder are updated using an asynchronous strategy; wherein, the state of the quantum bit is adjusted based on the restrictive optimization strategy.

Specifically, the parameters of the autoencoder network are initialized according to the requirements of the underlying quantum coding, including the weights and biases in the encoder and decoder, as well as the initialization state of the qubits.

Specifically, the encoder weight is initialized as follows:

Wherein, l is the level, σ ² is the initialization variance. In some preferred embodiments of the present invention, σ ² is set to 0.01.

And, the decoder weight is initialized as formula (22):

Moreover, the biases are initialized to zero, which can be expressed as formulas (23) and (24):

Furthermore, for qubit initialization, the initialization method of the potential quantum coding state can be expressed as formula (25):

Wherein, n is the number of qubits. In some preferred embodiments of the present invention, the number of qubits n is set to 8.

Furthermore, the data input into the dimensionality reduction model is forward propagated through the encoder, and the data is encoded into the latent space after layer-by-layer transformation. In this process, the data is compressed into a lower-dimensional representation, and the superposition and entanglement properties of quantum bits are used to enhance the feature expression ability of the data.

Specifically, the parameter calculation of the encoder forward propagation can be expressed as formula (26):

Among them, eX is the input data.

And, the activation function of the decoder forward propagation can be expressed as formula (27):

Among them, Re() is the ReLU activation function.

Furthermore, the latent space mapping method can be expressed as formula (28):

Among them, φ() is the quantum coding mapping function, and L is the last layer of the encoder.

In some preferred embodiments of the present invention, the mapping function The encoded feature vector is converted into the representation of quantum bits, and the calculation method can be expressed as formula (29):

Among them, eW _φ is the weight mapped to the latent quantum space, eb _φ is the bias mapped to the latent quantum space, and the softmax function is used to convert the output of the encoder into a probability distribution, which helps to represent the quantum bit state of the probability amplitude.

Furthermore, in the latent space, a restrictive optimization strategy is introduced to adjust the state of the quantum bit to ensure that key features are effectively retained while compressing the data.

Specifically, the latent space optimization method can be expressed as formula (30):

eL _quantum =Ψ(eQ _mapped ,eTargets) (30)

Among them, Ψ() is the optimization function of the quantum state, and eTargets is the training target, which is used to adjust the qubit to minimize the reconstruction error.

In some preferred embodiments of the present invention, the optimization function Ψ(eQ _mapped , eTargets) of the quantum state is responsible for adjusting the qubit to minimize the reconstruction error, which is achieved by introducing a quantum state similarity measurement. This embodiment uses the quantum Jensen-Shannon divergence for calculation, which can be expressed as formula (31):

Ψ(eQ _mapped ,eTargets)＝QJSD(eQ _mapped ∥eTargets) (31)

Among them, the QJSD function measures the similarity between the quantum state eQ _mapped and the target state eTargets.

Furthermore, the data in the latent space is reconstructed back to the original data space through the decoder, the reconstruction error is calculated, and the network parameters are updated according to the error through the back propagation algorithm. The present invention adopts an asynchronous strategy to update the parameters of the encoder and decoder to improve the training efficiency.

Specifically, in the process of back propagation and parameter updating, the reconstruction error calculation method can be expressed as formulas (32), (33) and (34):

Among them, L _dec is the last layer of the decoder, and Sig() is the Sigmoid activation function.

Furthermore, the parameter updating method can be expressed as formulas (35) and (36):

Among them, _ηm is the learning rate of the autoencoder.

Furthermore, the gradient and The specific calculation method can be expressed as formula (37):

in, The calculation of follows a similar chain rule and is performed from the decoder's perspective. In some preferred embodiments of the present invention, the learning rate η _m is set to 0.001.

The iterative training is repeated until the preset number of iterations is reached.

In some preferred embodiments of the present invention, the method of updating the bias is similar.

The data after feature extraction is input into the feature dimensionality reduction model for training. The embodiment of the present invention proposes an autoencoder neural network algorithm based on potential quantum coding to perform feature dimensionality reduction. Compared with the traditional autoencoder structure, the embodiment of the present invention introduces quantum bit representation to encode the latent space, and uses the superposition and entanglement characteristics of quantum bits to enhance the ability of feature expression, thereby achieving more efficient data compression and feature representation. In addition, the embodiment of the present invention guides the learning process of the latent space by introducing restrictive conditions, ensuring that the reconstructed data can remain highly similar to the original data while retaining more key information in the reduced data.

Through the above steps A1 and A2, we obtain a training data set that contains important features of the original data and has undergone effective dimensionality reduction. This data set provides a solid foundation for subsequent classifier training and helps improve the performance and accuracy of the classification model.

Furthermore, the existing model training methods may not be flexible enough in terms of learning rate adjustment and model optimization, which affects the training efficiency and classification accuracy of the model. In some preferred embodiments of the present invention, inputting the training data into the classifier to obtain the final classification model includes steps D1 to D2:

Step D1, divide the training data into a training set and a validation set.

Specifically, before training the classification model, the training data must first be split into a training set and a validation set. The training set is used to actually train the classifier, while the validation set is used to verify the performance of the model during the training process. The split ratio can be determined based on the specific situation. Common ratios include 70% training set and 30% validation set, or more complex split strategies such as cross-validation can be used. The purpose of splitting the data is to evaluate the generalization ability of the model during the training process, prevent overfitting, and ensure that the model can make accurate predictions on unseen data.

Step D2 trains the classifier based on the training set, and verifies the trained classifier based on the verification set until the loss function corresponding to the training set reaches a preset threshold, and determines the trained classifier as the final classification model; wherein, during the training process, the learning rate of the classifier is adjusted based on an adaptive learning rate adjustment mechanism.

Specifically, the classifier is trained using the data in the training set. The classifier can be any type of machine learning model, such as decision trees, support vector machines, neural networks, etc. The specific choice depends on the nature of the problem and the characteristics of the data. During the training process, the validation set is used to verify the performance of the classifier. This is usually done after each training iteration or several iterations to monitor the performance of the classifier on unknown data. Training continues until the loss function corresponding to the training set reaches a preset threshold. The loss function is an indicator that measures the difference between the model prediction and the actual label. When the loss is reduced to a sufficiently low level, it can be considered that the model has learned an effective pattern. During the training process, an adaptive learning rate adjustment mechanism is used to adjust the learning rate of the classifier. This mechanism allows the learning rate to be dynamically adjusted according to the performance of the model during the training process, which helps to increase the convergence speed and reduce fluctuations. Once the stopping condition is met, such as the loss function reaches a preset threshold or the accuracy on the validation set no longer improves, the trained classifier is determined as the final classification model.

In some preferred embodiments of the present invention, the training process of the probabilistic neural network classification model based on adaptive learning rate adjustment is as follows:

Initialize the weights and biases of the probabilistic neural network. Let p _θ represent the parameter set of the probabilistic neural network. The initialization process can be expressed as The subscript 0 refers to the initial state.

Set up an adaptive learning rate mechanism to dynamically adjust the learning rate based on the model's performance on the validation set. If the model performance improves, increase the learning rate to speed up convergence; if the model performance decreases, decrease the learning rate to avoid overfitting.

During the training process, by considering the manifold structure of the data in the high-dimensional space, the training strategy of the probabilistic neural network is adjusted to better capture the intrinsic characteristics of the data. Specifically, let the input feature vector be p _x , and transform it to the low-dimensional manifold space through the manifold mapping function M(p _x ). The mapping process can be expressed as formula (38):

M(p _x )=f _manifold (p _x ; p _φ ) (38)

Among them, p _φ is the parameter of the manifold mapping function.

In some preferred embodiments of the present invention, the mapping function f _manifold () is defined in detail using an autoencoder structure, assuming that the encoding part of the autoencoder is composed of the function Indicates that the decoding part is composed of the function Representation, where z is the encoded low-dimensional representation, is the parameter set for the encoding part, Parameter set for the decoding part.

Furthermore, the calculation method of the mapping function f _manifold (p _x ; p _φ ) can be expressed as formula (39):

Furthermore, repeated iterative training is performed. In each iteration, the model parameters are updated according to the current learning rate, and the performance of the model on the training set and the validation set is evaluated. This process is repeated until the performance of the model on the validation set is no longer significantly improved, or the preset maximum number of iterations is reached.

Specifically, in the manifold space, the training goal of the probabilistic neural network is to minimize the loss function L _PLNN , which can be expressed as formula (40):

Among them, p _y is the true label, is the label probability predicted by the probabilistic neural network, and i is the sample index.

Furthermore, the performance of the training process is monitored through the restart strategy. If the performance is lower than the preset threshold, the update strategy of the learning rate η and the parameter p _θ is dynamically adjusted, which is specifically expressed as formulas (41) and (42):

η ^(t+1) ＝η ^(t) ·γ _restart (41)

Among them, γ _restart is the factor for adjusting the learning rate, which is usually less than 1; is the gradient of the loss function.

In some preferred embodiments of the present invention, the performance is evaluated by using the accuracy on the validation set as an evaluation indicator.

The model with the best performance on the validation set is selected as the final model for the classification task of new samples. The final model output is formula (43):

Among them, softmax() is the Softmax function, which is used to convert the output of the neural network into a probability distribution.

In some preferred embodiments of the present invention, the predicted label probability It is based on the features after manifold mapping and the probabilistic neural network parameters p _θ . This process is implemented through the Softmax layer. Specifically, let h(z _i ; p _θ ) be the output layer of the probabilistic neural network. After receiving the mapped feature z _i , the output layer calculates a linear transformation, that is, formula (44):

h(z _i ; p _θ )=Wz _i +b (44)

Where W and b are the weights and biases of the output layer of the probabilistic neural network.

Furthermore, the predicted probability is calculated by applying the Softmax function and can be expressed as formula (45):

Through the above steps D1 and D2, we obtained a fully trained and validated classification model that can accurately predict the contamination level of soil samples. This machine learning process that combines data preprocessing, feature extraction, dimensionality reduction, and adaptive learning rate adjustment not only improves the performance of the model, but also ensures the stability and reliability of the model.

The reduced-dimensional data is input into a classifier. An embodiment of the present invention proposes a probabilistic neural network classification model based on adaptive learning rate adjustment. Different from a traditional probabilistic neural network, the probabilistic neural network based on adaptive learning rate adjustment not only considers the manifold structure of the data, but also introduces an adaptive learning rate adjustment mechanism, which can dynamically adjust the learning rate according to the performance of the model during the training process, thereby optimizing the model training process, improving classification accuracy, and accelerating convergence speed.

The embodiment of the present invention provides a soil contamination detection method based on artificial intelligence, which is based on the data expansion of the generative adversarial network of variational quantum coding, and utilizes the advantages of quantum computing to enhance the ability of the generative model. In addition, a high-dimensional latent space is constructed by a variational quantum algorithm to provide a rich data representation basis for data generation; the authenticity and diversity of data expansion are improved, and the generative model enhanced by quantum computing can generate more real and diverse data, which helps to improve the generalization ability and accuracy of the model. The neural network algorithm optimized by the policy-constrained submodular function is applied to data feature extraction, and the policy constraints are introduced to optimize the feature extraction process to ensure that the extracted features are both efficient and can retain the core information of the data. In addition, sparsity constraints are introduced in the feature extraction process to reduce the dimension of the feature space and improve the interpretability of the model; the neural network algorithm optimized by the policy-constrained submodular function can effectively extract features that are highly relevant to the soil contamination detection task and avoid interference from irrelevant features. The autoencoder based on potential quantum coding performs feature dimensionality reduction, enhances the feature expression capability through the superposition and entanglement characteristics of quantum bits, and achieves more efficient data compression and feature representation; the data compression efficiency is improved, and the autoencoder based on potential quantum coding can retain more key information in the dimensionality reduction process to achieve efficient data compression. A probabilistic neural network classification model based on adaptive learning rate adjustment introduces an adaptive learning rate adjustment mechanism to dynamically adjust the learning rate according to the performance of the model during training and optimize the model training process. The training process is optimized and the classification accuracy is improved. The adaptive learning rate adjustment mechanism can dynamically adjust the learning rate according to the model performance, accelerate the convergence speed and improve the classification accuracy.

Embodiment 3

On the basis of the above embodiments, an embodiment of the present invention provides a soil contamination detection device based on artificial intelligence. Referring to FIG3 , a structural diagram of a soil contamination detection device based on artificial intelligence provided by an embodiment of the present invention, the device includes:

The training data acquisition module 310 is used to acquire sample data of sample soil and label the sample data; wherein the labeled category represents the pollution degree of the sample soil.

The data expansion module 320 is used to perform data expansion on the labeled sample data through a generative adversarial network based on variational quantum coding to obtain expanded data.

The data processing module 330 is used to process the expanded data to obtain training data; wherein the data processing includes: feature extraction processing and feature dimension reduction processing.

The data classification module 340 is used to input the training data into the classifier to obtain the final classification model; wherein the classifier is a probabilistic neural network classifier based on adaptive learning rate adjustment.

The target data acquisition module 350 is used to acquire the target data corresponding to the target soil, and then process the target data to obtain the target data after feature reduction.

The data evaluation module 360 is used to input the target data after feature reduction into the final classification model to obtain the pollution degree corresponding to the target soil.

In some preferred embodiments of the present invention, the data expansion module 320 is used to construct a first quantum latent space by variable component quantum coding; the generator of the generative adversarial network generates data based on the features collected in the first quantum latent space; the discriminator of the generative adversarial network distinguishes the difference between the data generated by the generator and the real data; and the parameters of the generator and the parameters of the discriminator are iteratively optimized through adversarial training until the diversity of the data generated by the generator meets the preset diversity conditions.

In some preferred embodiments of the present invention, data is generated based on an incremental loss function constraint in adversarial training; wherein the incremental loss function is based on the loss function of the generator, a first adjustment coefficient of the penalty term in the iteration process, and a second adjustment coefficient constraint of the penalty term in the iteration process.

In some preferred embodiments of the present invention, the diversity of the data generated by the generator is evaluated based on the generated data, the average value of the generated data, the number of generated data, and the dimension of the generated data.

In some preferred embodiments of the present invention, the data processing module 330 is used to extract features from the expanded data through a neural network model based on submodular function optimization to obtain feature-extracted data; wherein, the optimization process of the submodular function is based on strategy constraints; the feature-extracted data is input into a feature dimensionality reduction model to obtain reduced data, and the reduced data is determined as training data; wherein, the feature dimensionality reduction model is determined based on an autoencoding neural network algorithm of potential quantum coding.

In some preferred embodiments of the present invention, the data processing module 330 is used to initialize the parameters of the neural network model; set the strategy constraints in the optimization process; wherein the strategy constraint function of the strategy constraint is based on the number of constraints, the weight corresponding to each constraint and the calculation formula constraints of each constraint; iterate the training operation until the preset maximum number of iterations is reached; the training operation includes: extracting features of the input data based on the neural network model; wherein the objective function of the feature extraction is based on the loss function of the input data, the sparsity constraint term, the hyperparameter for adjusting the sparsity and the hyperparameter constraint for the influence of the strategy constraint; evaluate the validity of the data after feature extraction, and adjust the strategy constraints and the parameters of the neural network model based on the evaluation results; wherein the evaluation function is based on the number of evaluation indicator points and the calculation formula constraints of the evaluation indicators.

In some preferred embodiments of the present invention, the data processing module 330 is used to adjust the parameters of the neural network model based on the dynamic feature-aware network adjustment mechanism; wherein the adjustment function is based on the current parameters of the neural network model, the gradient of the evaluation function relative to the current parameters, the network structure adjustment function, the learning rate and the dynamic adjustment factor constraints; the gradient of the evaluation function relative to the current parameters is based on the extracted eigenvalues, the target eigenvalues and the total number of features constraints; the network structure adjustment function is based on the hyperparameters for controlling the network adjustment amplitude, the hyperparameters for controlling the network sensitivity and the optimization direction constraints for the generator.

In some preferred embodiments of the present invention, the data processing module 330 is used to initialize the parameters of the feature dimensionality reduction model; iteratively perform the data dimensionality reduction operation until a preset maximum number of iterations is reached; the data dimensionality reduction operation includes: forwarding the original data of the input feature dimensionality reduction model through the encoder to the second quantum latent space to obtain the quantum bit data; based on the decoder, the quantum bit data is returned to the original space to obtain reconstructed data; based on the reconstructed data and the original data, the parameters of the encoder and the parameters of the decoder are updated using an asynchronous strategy; wherein, the state of the quantum bit is adjusted based on a restrictive optimization strategy.

In some preferred embodiments of the present invention, the data classification module 340 is used to divide the training data into a training set and a validation set; train the classifier based on the training set, and validate the trained classifier based on the validation set until the loss function corresponding to the training set reaches a preset threshold, and determine the trained classifier as the final classification model; wherein, during the training process, the learning rate of the classifier is adjusted based on an adaptive learning rate adjustment mechanism.

Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the soil contamination detection device based on artificial intelligence described above can refer to the corresponding process in the embodiment of the aforementioned soil contamination detection method based on artificial intelligence, and will not be repeated here.

Embodiment 4

An embodiment of the present invention further provides an electronic device for running a soil contamination detection method based on artificial intelligence; referring to FIG4 , a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention, the electronic device includes a memory 400 and a processor 401, wherein the memory 400 is used to store one or more computer instructions, and the one or more computer instructions are executed by the processor 401 to implement the above-mentioned soil contamination detection method based on artificial intelligence.

Furthermore, the electronic device shown in FIG. 4 further includes a bus 402 and a communication interface 403 , and the processor 401 , the communication interface 403 and the memory 400 are connected via the bus 402 .

The memory 400 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk storage. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 403 (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. may be used. The bus 402 may be an ISA bus, a PCI bus, or an EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one bidirectional arrow is used in FIG4, but it does not mean that there is only one bus or one type of bus.

The processor 401 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 401. The above processor 401 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in conjunction with the embodiments of the present invention can be directly embodied as a hardware decoding processor for execution, or a combination of hardware and software modules in the decoding processor for execution. The software module may be located in a storage medium mature in the art, such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 400, and the processor 401 reads the information in the memory 400 and completes the steps of the method of the above embodiment in combination with its hardware.

An embodiment of the present invention further provides a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement the above-mentioned business recommendation method. The specific implementation can be found in the method embodiment, which will not be repeated here.

The computer program product of the soil contamination detection method, device and electronic device based on artificial intelligence provided in the embodiments of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the method in the previous method embodiment. The specific implementation can be referred to the method embodiment, which will not be repeated here.

Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system and/or device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

In addition, in the description of the embodiments of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims (10)

1. A soil contamination detection method based on artificial intelligence, characterized by comprising: Acquire sample data of the sample soil, and label the sample data; wherein the labelled category represents the degree of contamination of the sample soil; Performing data expansion on the labeled sample data through a generative adversarial network based on variational quantum coding to obtain expanded data; Processing the expanded data to obtain training data; wherein the data processing includes: feature extraction processing and feature dimension reduction processing; Inputting the training data into a classifier to obtain a final classification model; wherein the classifier is a probabilistic neural network classifier based on adaptive learning rate adjustment; After acquiring the target data corresponding to the target soil, the target data is subjected to the data processing to obtain the target data after feature reduction; The target data after feature reduction is input into the final classification model to obtain the pollution degree corresponding to the target soil. 2. The soil contamination detection method based on artificial intelligence according to claim 1 is characterized in that the step of training the generative adversarial network based on variational quantum coding comprises: Constructing the first quantum potential space through variable component quantum coding; The generator of the generative adversarial network generates data based on features collected in the first quantum latent space; The discriminator of the generative adversarial network distinguishes the difference between the data generated by the generator and the real data; The parameters of the generator and the parameters of the discriminator are iteratively optimized through adversarial training until the diversity of the data generated by the generator meets the preset diversity conditions. 3. According to the artificial intelligence-based soil contamination detection method of claim 2, it is characterized in that data is generated based on an incremental loss function constraint in the adversarial training; wherein the incremental loss function is based on the loss function of the generator, the first adjustment coefficient of the penalty term in the iterative process, and the second adjustment coefficient constraint of the penalty term in the iterative process. 4. The artificial intelligence-based soil contamination detection method according to claim 2 is characterized in that the diversity of the data generated by the generator is based on the generated data, the average value of the generated data, the number of the generated data and the dimensional evaluation of the generated data. 5. The soil contamination detection method based on artificial intelligence according to claim 1 is characterized in that the step of processing the expanded data to obtain training data comprises: Extracting features from the expanded data through a neural network model optimized based on a submodular function to obtain feature-extracted data; wherein the optimization process of the submodular function is constrained based on a strategy; The feature-extracted data is input into a feature dimension reduction model to obtain dimension-reduced data, and the dimension-reduced data is determined as the training data; wherein the feature dimension reduction model is determined based on a self-encoding neural network algorithm of potential quantum coding. 6. The soil contamination detection method based on artificial intelligence according to claim 5 is characterized in that the step of training the neural network model based on submodular function optimization comprises: Initializing parameters of the neural network model; Setting strategy constraints in the optimization process; wherein the strategy constraint function of the strategy constraint is based on the number of constraints, the weight corresponding to each constraint, and the calculation formula constraint of each constraint; The training operation is iterated until a preset maximum number of iterations is reached; the training operation includes: extracting features from the input data based on the neural network model; wherein the objective function of the feature extraction is based on the loss function of the input data, the sparsity constraint term, the hyperparameter for adjusting the sparsity, and the hyperparameter constraint for the influence of the strategy constraint; evaluating the validity of the data after the feature extraction, and adjusting the strategy constraint conditions and the parameters of the neural network model based on the evaluation results; wherein the evaluation function is based on the number of evaluation index points and the calculation formula constraints of the evaluation index. 7. The soil contamination detection method based on artificial intelligence according to claim 6 is characterized in that the step of adjusting the parameters of the neural network model based on the evaluation results comprises: The parameters of the neural network model are adjusted based on a dynamic feature-aware network adjustment mechanism; wherein the adjustment function is based on the current parameters of the neural network model, the gradient of the evaluation function relative to the current parameters, the network structure adjustment function, the learning rate and the dynamic adjustment factor constraints; the gradient of the evaluation function relative to the current parameters is based on the extracted eigenvalues, the target eigenvalues and the total number of features constraints; the network structure adjustment function is based on the hyperparameters for controlling the network adjustment amplitude, the hyperparameters for controlling the network sensitivity and the optimization direction constraints for the generator. 8. The soil contamination detection method based on artificial intelligence according to claim 5 is characterized in that the step of training the feature dimension reduction model comprises: Initializing parameters of the feature dimensionality reduction model; Iteratively perform a data dimension reduction operation until a preset maximum number of iterations is reached; the data reduction operation includes: forward propagating the original data input into the feature dimension reduction model to a second quantum potential space through an encoder to obtain quantum bit data; based on a decoder, the quantum bit data is transferred to the original space to obtain reconstructed data; based on the reconstructed data and the original data, an asynchronous strategy is used to update the parameters of the encoder and the decoder; wherein, the state of the quantum bit is adjusted based on a restrictive optimization strategy. 9. The soil contamination detection method based on artificial intelligence according to claim 1 is characterized in that the step of inputting the training data into the classifier to obtain the final classification model comprises: Dividing the training data into a training set and a validation set; The classifier is trained based on the training set, and the trained classifier is verified based on the verification set until the loss function corresponding to the training set reaches a preset threshold, and the trained classifier is determined as the final classification model; wherein, during the training process, the learning rate of the classifier is adjusted based on an adaptive learning rate adjustment mechanism. 10. A soil contamination detection device based on artificial intelligence, characterized by comprising: A training data acquisition module is used to acquire sample data of sample soil and to label the sample data; wherein the labeled category represents the degree of contamination of the sample soil; A data expansion module, used to perform data expansion on the labeled sample data through a generative adversarial network based on variational quantum coding to obtain expanded data; A data processing module is used to process the expanded data to obtain training data; wherein the data processing includes: feature extraction processing and feature dimension reduction processing; A data classification module, used for inputting the training data into a classifier to obtain a final classification model; wherein the classifier is a probabilistic neural network classifier based on adaptive learning rate adjustment; A target data acquisition module, used for acquiring target data corresponding to the target soil, and then performing the data processing on the target data to obtain target data after feature reduction; The data evaluation module is used to input the target data after feature reduction into the final classification model to obtain the pollution degree corresponding to the target soil.