CN117853260A

CN117853260A - Crop yield prediction method and system based on multi-source heterogeneous data

Info

Publication number: CN117853260A
Application number: CN202410158291.0A
Authority: CN
Inventors: 何岩峰; 杜纪龙; 李新峰; 罗素华; 于鑫; 王燕娜
Original assignee: Zhongdian Yunke Information Technology Co ltd
Current assignee: Zhongdian Yunke Information Technology Co ltd
Priority date: 2024-02-04
Filing date: 2024-02-04
Publication date: 2024-04-09

Abstract

The invention discloses a crop growth situation monitoring method based on multisource heterogeneous data, which comprises the following steps: acquiring multi-source heterogeneous data and a multi-spectrum remote sensing image in a crop growth period, wherein the multi-source heterogeneous data at least comprises soil data, surrounding environment data and crop data; constructing a depth space spectrum feature learning module based on a three-dimensional convolutional neural network; constructing a time correlation modeling module through a long-term and short-term memory network, wherein the time correlation modeling module is used for extracting time information in multispectral remote sensing images or multisource heterogeneous data; building a growth situation awareness model based on the depth space spectrum feature learning module and the time correlation modeling module; the growth situation awareness model can accurately predict the yield of the large-area crops.

Description

Crop yield prediction method and system based on multi-source heterogeneous data

Technical Field

The invention belongs to the technical field of agricultural informatization, and particularly relates to a crop yield prediction method and system based on multi-source heterogeneous data.

Background

In recent years, world agricultural production faces a series of challenges. With single farming agriculture, double-season farming, crop rotation and insufficient soil rest time becoming more and more common, traditional extensive and high-input planting systems and agricultural technologies have gradually reached efficacy limits, as a traditional agricultural large country, agricultural production occupies a significant position in economic development of China, and crop industry development is directly related to national grain safety and social stability. Strengthening grain safety guarantee is a great national demand, developing intelligent agriculture, improving scientificity, precision and intelligence of grain production technical service and management decision is a fundamental way for guaranteeing grain safety, and objectively and accurately analyzing growth situation of large-area crops is an important problem to be solved in intelligent agriculture.

In recent years, researchers continuously conduct intensive research on crop modeling and growth situation rules, or dynamically simulate the natural growth process of real plants from physiological ecology based on rules; or performing three-dimensional reconstruction of plants by using a machine vision algorithm based on the images; or quantifying and modeling geometric parameters of the plant based on the point cloud data obtained by laser scanning. However, studies have been focused mainly on the growth analysis of single plants, the growth analysis of small areas, or the influence of single type data on plant growth situation, and rarely involve an important study topic of population growth situation analysis of large-area crops based on multi-source data.

Therefore, aiming at the problems, how to provide a method for acquiring crop growth situation information from complex data, so that crop modeling is improved from pure single plant situation calculation to large-area group growth situation, and accurate yield prediction is carried out on large-area crops, and the method is a technical problem to be solved at present.

Disclosure of Invention

The invention aims to provide a crop yield prediction method and system based on multi-source heterogeneous data, which are used for solving the problems in the prior art.

In order to achieve the above purpose, the present invention adopts the following technical scheme:

a crop yield prediction method based on multi-source heterogeneous data, comprising:

acquiring multi-source heterogeneous data and a multi-spectrum remote sensing image in a crop growth period, wherein the multi-source heterogeneous data at least comprises soil data, surrounding environment data and crop data;

constructing a depth space spectrum feature learning module based on a three-dimensional convolutional neural network;

constructing a time correlation modeling module through a long-term and short-term memory network, wherein the time correlation modeling module is used for extracting time information in multispectral remote sensing images or multisource heterogeneous data;

building a growth situation awareness model based on the depth space spectrum feature learning module and the time correlation modeling module;

and predicting the yield of the crops according to the growth situation awareness model.

In some embodiments, after constructing the deep space spectrum feature learning module based on the three-dimensional convolutional neural network, the method further includes:

and converting the multispectral remote sensing image or the multisource heterogeneous data into a series of high-level spatial spectrum features through a deep spatial spectrum feature learning module.

In some embodiments, converting the multispectral remote sensing image into a series of advanced spatial spectral features by the deep spatial spectral feature learning module comprises:

defining the acquired multispectral remote sensing image as V _i ∈R ^h×w×c×T Wherein R represents a real number set, h×w represents the size of an image, c represents the number of spectrum channels, and T represents the number of the spectrum channels of cropsThe number of multispectral remote sensing images obtained in the growth period;

splitting the multispectral remote sensing images based on the number of the multispectral remote sensing images obtained under the time sequence to obtain a plurality of remote sensing image blocks X;

for remote sensing image block at t momentAfter the three-dimensional convolution layer slides the three-dimensional convolution kernels along the wide height and time dimensions of the remote sensing image block, each three-dimensional convolution kernel generates a three-dimensional feature map for each space-spectrum-time data volume, and the three-dimensional feature map is determined according to the following formula (1):

wherein F is _C3D (w, h, b) represents a series of high-level spatial spectral features,representing three-dimensional convolution operation, w ^pqr Representing the weight matrix in the three-dimensional convolution kernel, p, q, r representing the element index of the convolution kernel.

In some embodiments, the long-term and short-term memory network models the time characteristics of crops based on a three-dimensional convolutional neural network, and uses the time sequence spatial spectrum characteristics F= { F ₁ ,F ₂ …F _t And the model is input to obtain a time correlation modeling module, and the long-term and short-term memory network comprises three gating units, wherein the three gating units comprise an input gate, a forget gate and an output gate.

In some embodiments, three gating cells are obtained according to the following formulas (2), (3) and (4), including:

wherein F is _t Indicating forgetful door, I _t Represents the input gate, O _t Representing the output gate, σ represents the Sigmoid activation function, w _f 、w _i And w _o The weight matrix is represented by a matrix of weights,input data representing the current time step, b _f 、b _i And b _o Represents the offset value, h _t-1 Indicating the hidden state of the last moment.

In some embodiments, further comprising a cellular state and a hidden state, the cellular state determining a next hidden state using an output gate, the cellular state and the hidden state being obtained according to the following formulas (5) and (6):

h _t ＝O _t ×tanh (C _t ) (6)

wherein, I _t Input representing the current time step, O _t Representing output, C _t Representing the current cell state, C _t-1 Representing the cell state of the model output at the previous moment, h _t Indicating the hidden state, b _c Representing the bias value, tanh is the activation function.

In some embodiments, a full connection layer is connected after the time correlation modeling module to conceal the state h of the last time step _t Mapping to yield values of crops, wherein a complete growth situation perception model is constructed after a full-connection layer is connected behind a time correlation modeling module.

In some embodiments, acquiring multi-source heterogeneous data over a crop growth cycle includes:

acquiring soil data and surrounding environment data based on the sensor equipment of the Internet of things, wherein the soil data at least comprises soil temperature, volume water content, photosynthetic radiation, soil water potential and soil oxygen, and the surrounding environment data at least comprises precipitation, evaporation capacity, temperature, humidity, air pressure, wind speed condition, wind direction condition, illumination, total radiation of CO2 content and photosynthetic effective radiation;

crop data is acquired based on the intelligent vegetation physiological sensor device, and the crop data at least comprises leaf surface humidity and leaf surface temperature.

In some embodiments, after obtaining the multi-source heterogeneous data over the crop growth cycle, further comprising:

setting the value range of each multi-source heterogeneous data, and carrying out alarm processing if the value range exceeds the preset value range.

Correspondingly, the application also provides a crop yield prediction system based on the multi-source heterogeneous data, which comprises the following steps:

the acquisition module is used for acquiring multi-source heterogeneous data and multi-spectrum remote sensing images in a crop growth period, wherein the multi-source heterogeneous data at least comprise soil data, surrounding environment data and crop data;

the first construction module is used for constructing a depth space spectrum feature learning module based on the three-dimensional convolutional neural network;

the second construction module is used for constructing a time correlation modeling module through a long-term and short-term memory network, and the time correlation modeling module is used for extracting time information in the multispectral remote sensing image or the multisource heterogeneous data;

the third construction module is used for constructing a growth situation awareness model based on the depth space spectrum feature learning module and the time correlation modeling module;

and the yield prediction module is used for predicting the yield of the crops according to the growth situation awareness model.

The beneficial effects are that:

1) The method is applied to large-area crop growth monitoring, crop abnormality or health problems are found early, and monitoring and insight level of farms are enhanced; identifying and monitoring plant growth condition changes and distribution conditions of different growth conditions, and carrying out differential management; monitoring the influence of measures such as fertilization, irrigation, pesticide spraying and the like on crop health of specific tasks; the capacity of the current year is predicted by benchmark capacity analysis of different areas and years.

2) The application of the application in the agricultural field can lead the agricultural planting to be more scientific, the disease and insect prevention to be more specialized, reduce the harm of chemical agents such as pesticides and the like to the environment, protect the ecological environment and promote sustainable development; on the other hand, the agricultural resources are reasonably utilized, the growth vigor of crops is ensured, the yield and quality of the crops are improved, meanwhile, the consumption of manpower and material resources is reduced, the labor productivity is improved, the cost investment is reduced, and huge economic and ecological benefits are brought.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this application, are included to provide a further understanding of the application and to provide a further understanding of the application with regard to the other features, objects and advantages of the application. The drawings of the illustrative embodiments of the present application and their descriptions are for the purpose of illustrating the present application and are not to be construed as unduly limiting the present application. In the drawings:

fig. 1 is a schematic flow chart of a crop yield prediction method based on multi-source heterogeneous data according to an embodiment of the present application; the description of the second figure should be omitted here

Fig. 2 is a schematic structural diagram of a crop growth situation awareness model according to an embodiment of the present application;

FIG. 3 is a schematic structural diagram of a time-dependent modeling module according to an embodiment of the present application;

fig. 4 is a schematic structural diagram of a crop yield prediction system based on multi-source heterogeneous data according to an embodiment of the present application.

Detailed Description

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the present invention will be briefly described below with reference to the accompanying drawings and the description of the embodiments or the prior art, and it is obvious that the following description of the structure of the drawings is only some embodiments of the present invention, and other drawings can be obtained according to these drawings without inventive effort to a person skilled in the art. It should be noted that the description of these examples is for aiding in understanding the present invention, but is not intended to limit the present invention.

It will be understood that, although the terms first, second, etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another element. For example, a first element could be termed a second element, and, similarly, a second element could be termed a first element, without departing from the scope of example embodiments of the present invention.

It should be understood that for the term "and/or" that may appear herein, it is merely one association relationship that describes an associated object, meaning that there may be three relationships, e.g., a and/or B, may represent: a alone, B alone, and both a and B; for the term "/and" that may appear herein, which is descriptive of another associative object relationship, it means that there may be two relationships, e.g., a/and B, it may be expressed that: a alone, a alone and B alone; in addition, for the character "/" that may appear herein, it is generally indicated that the context associated object is an "or" relationship.

Example 1:

fig. 1 is a schematic flow chart of a crop yield prediction method based on multi-source heterogeneous data according to an embodiment of the present invention, including:

s201, multi-source heterogeneous data and multi-spectrum remote sensing images in a crop growth period are obtained.

Specifically, the acquisition of multi-source heterogeneous data is the basis of crop growth situation awareness, so that the application firstly acquires related data in various crop growth periods based on various Internet of things sensor devices and remote sensing satellites. In order to fully monitor the environmental conditions in the crop growth process, the production key elements such as soil, atmosphere, water and the like are intelligently acquired based on the sensor equipment of the Internet of things.

Specifically, soil is the most important factor in determining the growth situation of crops, and soil data and surrounding environment data are acquired based on the sensor device of the internet of things, so that the soil data are acquired first: various soil metrics including soil temperature, volumetric water content, photosynthetic radiation, soil water potential and soil oxygen level. In addition, the moisture in the soil is an important nutrient for all crops and plants, and the yield of crops is greatly affected by the supply of moisture in the soil. Therefore, key vegetation growth indexes such as leaf surface humidity, leaf surface temperature and the like are continuously obtained based on intelligent vegetation physiological sensor equipment. Meanwhile, in order to fully consider the influence of the atmospheric environment on the growth of crops, indexes such as precipitation, temperature, humidity, air pressure, wind speed, wind direction and the like are further collected. Finally, in order to realize long-time and large-range intelligent perception of the growth of crops, multispectral remote sensing images in the growth period of the crops are further acquired based on the sentinel satellite data.

In general, in order to fully sense the change of the crop growth process, the application provides more than 20 internet of things sensor data including { { soil temperature, volumetric water content, photosynthetic radiation, soil water potential, pH value, nitrogen, phosphorus, potassium }, { leaf surface humidity, leaf surface temperature }, { precipitation, evaporation, temperature, humidity, air pressure, wind speed, wind direction, illumination, total radiation of CO2 content, photosynthetic effective radiation } }, and long-time sequence multispectral satellite image data.

S202, constructing a deep space spectrum feature learning module based on a three-dimensional convolutional neural network.

And a depth space-spectrum characteristic learning module for constructing a growth situation awareness model based on a three-dimensional convolutional neural network (3D CNN) is used for deeply excavating the growth characteristics of crops. The depth space-spectrum characteristic learning module is formed by stacking 5 three-dimensional convolution layers, 3 pooling layers, 1 full-connection layer, a series of three-dimensional batch normalization layers and an activation function. Wherein the convolution layer is used for extracting features from the data; adding a three-dimensional batch normalization layer and a ReLU activation function after each convolution layer for accelerating model convergence; the pooling layer is used for reducing model parameters and feature dimension reduction; finally, a full connection layer is connected to map the features extracted by the pooling layer to an output space for preparing for subsequent tasks.

In order to realize accurate prediction of crop growth situation, in the preferred embodiment of the present scheme, after constructing the deep space spectrum feature learning module based on the three-dimensional convolutional neural network, the method further includes:

and converting the multispectral remote sensing image or the related data into a series of high-level spatial spectrum features through a deep spatial spectrum feature learning module.

Specifically, the acquired multispectral remote sensing image is defined as V _i ∈R ^h×w×c×T Wherein R represents a real number set, h×w represents the size of an image, c represents the number of spectrum channels, and T represents the number of multispectral remote sensing images obtained by crops in a growth period;

wherein F is _C3D (w, h, b) represents a series of high-level spatial spectral features,representing three-dimensional convolution operation, w ^pqr And (3) representing a weight matrix in the three-dimensional convolution kernel, wherein p, q and r represent element indexes of the convolution kernel, and the three-dimensional convolution layer is a part of the inside of the depth space spectrum characteristic learning module.

S203, constructing a time correlation modeling module through a long-term and short-term memory network.

Specifically, the long-term and short-term memory network models the time characteristics of crops based on a three-dimensional convolutional neural network, and uses the time sequence spatial spectrum characteristics F= { F ₁ ,F ₂ …F _t And the model is input to obtain a time correlation modeling module, and the long-term and short-term memory network comprises three gating units, wherein the three gating units comprise an input gate, a forget gate and an output gate.

To obtain three gating units, in a preferred embodiment of the present solution, three gating units are used: input gate, forget gate and output gate to control the input, forget and output of information to more effectively handle long-term dependencies.

Three gating cells are obtained according to the following formulas (2), (3) and (4), including:

In order to achieve an accurate prediction of the yield value of the crop, in a preferred embodiment of the present solution, further comprising a cellular state and a concealment state, said cellular state determining the next concealment state using the output gate, the cellular state and the concealment state being obtained according to the following formulas (5) and (6):

h _t ＝O _t ×tanh (C _t ) (6)

And connecting a full-connection layer behind the time correlation modeling module, and mapping the hidden state of the last time step into a yield value of crops, wherein a complete growth situation perception model is constructed after connecting the full-connection layer behind the time correlation modeling module.

The application is applied to large-area crop growth monitoring, crop abnormality or health problems are found early, and monitoring and insight level of a farm are enhanced; identifying and monitoring plant growth condition changes and distribution conditions of different growth conditions, and carrying out differential management; monitoring the influence of measures such as fertilization, irrigation, pesticide spraying and the like on crop health of specific tasks; the capacity of the current year is predicted by benchmark capacity analysis of different areas and years.

Example 2

In order to fully excavate the growth rule of crops, the growth situation of crops is better perceived. The method further builds a growth situation perception model on the basis of the acquired long-time sequence remote sensing image data. And winter wheat and corn are used as target crops, so that the accurate prediction of the yield of the crops is realized. Specifically, an end-to-end deep learning joint network is constructed, and a multi-time phase-multi-spectrum remote sensing image is utilized, and simultaneously, temperature, precipitation data and the like are combined, so that a crop growth mode is identified from multiple dimensions and is used for crop yield prediction. The growth situation perception model can deeply excavate space-spectrum characteristics from the multispectral remote sensing image and is used for representing the growth situation of crops. On the basis, the inherent time correlation in the multi-time-phase image is further excavated, so that the time change rule of crop growth is better modeled. Meanwhile, through the future possible change of the set temperature or precipitation, the physiological situation change condition of crops under the external ecological environment change can be simulated. The crop growth situation awareness model is shown in figure 2.

2.1 crop depth space-spectrum feature learning module:

and a depth empty spectrum feature learning module for constructing a growth situation awareness model based on a three-dimensional convolutional neural network (3D CNN) is used for deeply excavating the growth features of crops. The depth space spectrum feature learning module is formed by stacking 5 three-dimensional convolution layers, 3 pooling layers, 1 full-connection layer, a series of three-dimensional batch normalization layers and an activation function. Wherein the convolution layer is used for extracting features from the data; adding a three-dimensional batch normalization layer and a ReLU activation function after each convolution layer for accelerating model convergence; the pooling layer is used for reducing model parameters and feature dimension reduction; finally, a full connection layer is connected to map the features extracted by the pooling layer to an output space for preparing for subsequent tasks.

Specifically, firstly, the input of a spatial spectrum feature learning module is standardized, and the acquired multispectral remote sensing image is defined as V _i ∈R ^h×w×c×T Where h w represents the image size, c represents the number of spectral channels, and T represents the time series of the crop. Resolving remote sensing images based on time sequences to obtain T sequences X= { X ₁ ,X ₂ …X _T As input of model, for remote sensing image block at t timeThe three-dimensional convolution layer slides three-dimensional convolution kernels along the wide height and time dimensions of the image, each convolution kernel generating a three-dimensional feature map for each space-spectrum-time data volume, as in (1):

wherein F is _C3D (w, h, b) represents a series of high-level spatial spectral features (i.e., a three-dimensional feature map),representing three-dimensional convolution operation, w ^pqr Representing the weight matrix in the three-dimensional convolution kernel, p, q, r representing the element index of the convolution kernel. The original multispectral time sequence image extracted by the deep space spectrum feature learning module is converted into a series of high-level space spectrum features F= { F ₁ ,F ₂ …F _T }。

2.2 crop growth time correlation modeling module:

in different climatic periods of crops, the plant morphology of the crops has obvious change, the time sequence information in the multi-temporal images is integrated, the correlation relationship of the growth of the crops is modeled, and the method is important for sensing the growth situation of the crops and realizing the prediction of the crop yield. The advanced spatial spectrum features extracted by the crop depth spatial-spectral feature learning module at each time step are further utilized to construct a time correlation modeling module to deeply mine time information in the multi-temporal images by utilizing a long-short-term memory network (LSTM), as shown in figure 2.

Specifically, LSTM models crop time characteristics based on Recurrent Neural Networks (RNNs) and introduces gating mechanisms through three gating units: input gate, forget gate and output gate to control the input, forget and output of information to more effectively handle long-term dependencies. In time sequence, spatial spectrum characteristic F= { F ₁ ,F ₂ …F _t The time sequence empty spectrum feature is the output of the depth empty spectrum feature learning module. h is a _t-1 Representing the hidden state at time t-1, three gate units in LSTM may be represented as:

the three gate units in LSTM can be expressed as:

h _t ＝O _t ×tanh (C _t ) (6)

in the above formula, sigma is a Sigmoid activation function, w _f 、w _i 、w _o As a matrix of weights, the weight matrix,input data representing the current time step, b _f 、b _i 、b _c 、b _o Represents the bias value, F _t 、I _t And O _t Distribution representing forget gate, input gate and output gate, C _t Representing the current cell state, C _t-1 Representing the cell state of the model output at the previous moment, h _t Indicating the hidden state, b _c Representing the bias value, tanh is the activation function.

The forget gate is used to determine which information needs to be deleted from a particular moment, and the input gate is used to calculate the current cell state C _t Determining a next hidden state h by using an output gate _t To make the final prediction. These gate units and cell states enable the LSTM structure to memorize long time series of information. Finally, connecting a full connection layer FC at the end of the module, and hiding the hidden state h of the last time step _t Mapped to a final yield value.

It should be noted that, in the embodiment of the present application, the obtained multispectral remote sensing image is used to predict the yield in a large area, and when the obtained multisource heterogeneous data is input into the growth situation model, different growth situation analyses can be performed according to the input data situation.

According to the intelligent monitoring system, the growth situation (small-area monitoring) of crops can be monitored by integrating multi-source heterogeneous data, and the intelligent monitoring system can further construct a group situation perception informatization platform integrating environment monitoring, crop model analysis and accurate regulation, so that intelligent guidance is provided for agricultural production.

Data dynamic monitoring: based on all kinds of thing networking sensor data and intelligent vegetation physiological sensor equipment, the change condition of external environment in the real-time supervision crops growth process mainly includes following module:

(1) Data storage and management: in order to effectively store various indexes, the platform firstly creates a monitoring index data table field and stores various indexes in a table form. And then, acquiring various index values related to the atmosphere, the water and the soil at regular time, and continuously storing the acquired information into a database.

(2) And (5) quickly inquiring index information: the internet of things element data of any time, last time, next time and latest time can be quickly checked, and the internet of things element data comprises elements such as real-time air temperature, wind direction and speed, rainfall, soil water content and the like. The live distribution map is drawn on the crop unit distribution in the research area, and the live data distribution condition in each unit can be clearly seen.

(3) Real-time early warning module: and setting the value range of each index according to the agricultural production environment standard. While monitoring the sensor data, the indexes exceeding the standard range are classified into different levels according to the size of the exceeding range, and labeling and alarming (red, orange, yellow and blue) are carried out. The alarm strategy can set starting strategies such as voice alarm, telephone alarm and the like, and set alarm timeliness and data delay time. And meanwhile, the data automatic refreshing is supported, the early warning icon is clicked, and the early warning detailed content is displayed.

3.1 visual analysis of growth situation

Under the support of real-time data, the platform is further supported by an ecological perception model, and visual analysis of crop growth situation is carried out aiming at winter wheat and corn. The device mainly comprises the following modules:

(1) Visualization of yield data: based on the model input values, the spatial distribution of the yield of winter wheat and corn is visually predicted in the form of a map. Meanwhile, according to the size of the corresponding yield value, the research area is divided to form three different levels of a yield high-value area, a yield medium-value area and a yield low-value area.

(2) Yield trend analysis: based on the annual yield prediction data, a graph of the time-series evolution of winter wheat and corn yields is graphically displayed. And meanwhile, the background automatically analyzes the evolution trend. And setting a region with a yield showing a decreasing trend as a focus region, marking the region by using red and issuing early warning information. And simultaneously supporting data click query and displaying various sensor index values of the area.

3.2 Intelligent external ecological data Regulation

Based on the crop ecological perception model, the intelligent regulation of external ecological data is further realized. Specifically, by setting different temperatures and precipitation conditions, the ecological perception model is utilized to respectively obtain the change conditions of crop yield under the conditions of temperature rise and fall of 0.1, 0.5 and 1 ℃ and the conditions of precipitation rise or fall of 5%,10% and 20%. And show the variation in yield in the form of a graph. Based on the method, the method is combined with a trend analysis method to analyze the condition of the yield, and the external ecological data is intelligently adjusted according to the setting of the condition of the yield increase.

Fig. 4 is a schematic structural diagram of a crop yield prediction system based on multi-source heterogeneous data, which includes the following steps:

the acquisition module 10 is configured to acquire multi-source heterogeneous data and a multi-spectrum remote sensing image in a crop growth period, where the multi-source heterogeneous data at least includes soil data, surrounding environment data and crop data;

a first construction module 20, configured to construct a deep space spectrum feature learning module based on a three-dimensional convolutional neural network;

the second construction module 30 is configured to construct a time correlation modeling module through a long-term and short-term memory network, where the time correlation modeling module is configured to extract time information in the multispectral remote sensing image or the multisource heterogeneous data;

a third construction module 40, configured to construct a growth situation awareness model based on the deep space spectrum feature learning module and the time correlation modeling module;

and the yield prediction module 50 is used for predicting the yield of the crops according to the growth situation awareness model.

Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

The foregoing description is only of the preferred embodiments of the invention and is not intended to limit the scope of the invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A crop yield prediction method based on multi-source heterogeneous data, comprising:

constructing a time correlation modeling module through a long-term and short-term memory network;

2. The method of claim 1, further comprising, after constructing the deep spatial feature learning module based on the three-dimensional convolutional neural network:

3. The method of claim 2, wherein the deep spatial feature learning module comprises at least a three-dimensional convolution layer, and converting the multispectral remote sensing image to a series of advanced spatial features by the deep spatial feature learning module comprises:

defining the acquired multispectral remote sensing image as V _i ∈R ^h×w×c×T Wherein R represents a real number set, h×w represents the size of an image, c represents the number of spectrum channels, and T represents the number of multispectral remote sensing images obtained by crops in a growth period;

wherein F is _C3D (w, h, b) represents a series of high-level spatial spectral features,representing three-dimensional convolution operation, w ^pqr Representing a three-dimensional rollThe weight matrix in the product kernel, p, q, r, represents the element index of the convolution kernel.

4. The method of claim 1, wherein the long-term and short-term memory network models crop time characteristics based on a three-dimensional convolutional neural network and uses time-series spatial spectral characteristics f= { F ₁ ,F ₂ …F _t And the time sequence empty spectrum characteristic is the output of the deep empty spectrum characteristic learning module.

5. The method of claim 4, wherein the obtaining three gating cells according to the following formulas (2), (3) and (4) comprises:

6. The method of claim 5, further comprising a cellular state and a hidden state, the cellular state determining a next hidden state using an output gate, the cellular state and the hidden state being obtained according to the following formulas (5) and (6):

h _t ＝O _t ×tanh(C _t ) (6)

wherein I is _t Input representing the current time step, O _t Representing output, C _t Representing the current cell state, C _t-1 Representing the cell state of the model output at the previous moment, h _t Indicating the hidden state, b _c Representing the bias value, tanh is the activation function.

7. The method of claim 6, wherein a fully connected layer is connected after the time-dependent modeling module, and the hidden state of the last time step is mapped to a yield value of the crop, wherein a complete growth situation awareness model is constructed after the fully connected layer is connected after the time-dependent modeling module.

8. The method of claim 1, wherein acquiring multi-source heterogeneous data over a crop growth cycle comprises:

9. The method of claim 8, further comprising, after obtaining the multi-source heterogeneous data over the crop growth cycle:

10. A crop yield prediction system based on multi-source heterogeneous data, comprising:

the second construction module is used for constructing a time correlation modeling module through the long-term and short-term memory network;