CN118135388A - An underwater target detection model and method based on image enhancement and AR mechanism - Google Patents

Abstract

The invention provides an underwater target detection model and method based on an image enhancement and AR mechanism, which belong to the technical field of underwater target image enhancement and detection. The use of the remaining attention structure allows the network to select valid feature information. The method combines residual feature attention blocks and novel combinations of multi-scale and multi-block structures. The multiple networks extract local features to accommodate various underwater images. The problems of low contrast, unclear edges, color distortion and the like of the original underwater image are solved. Then, detection of marine life based on visual attention and relational mechanisms employs a new approach to applying improved AR modules to high-efficiency marine life detectors that can well improve the recognition of life in complex underwater environments.

Description

An underwater target detection model and method based on image enhancement and AR mechanism

Technical Field

The present invention belongs to the technical field of underwater target image enhancement and detection, and in particular relates to an underwater target detection model and method based on image enhancement and AR mechanism.

Background technique

Underwater target detection is an important topic in underwater detection applications. As we all know, the ocean covers more than 70% of the earth's surface. The ocean provides humans with rich mineral resources, biological resources and marine food. The "China Marine Economic Bulletin" shows that my country's marine-related GDP exceeded 7.7 trillion in 2017, a year-on-year increase of 6.9%, and the marine GDP accounted for 9.4% of GDP. With the increasing value of marine resources, how to accurately detect and identify underwater targets has become a hot issue in the field of marine technology research.

Due to the complex scattering of underwater targets, low visibility, and high target clutter, underwater target detection faces huge challenges. Traditional imaging technology often fails to capture the details and surface features of underwater targets, and it is difficult to detect and identify the material characteristics of the target. Therefore, it is very necessary to perform image enhancement on underwater images. Existing underwater image enhancement algorithms can be divided into two categories: image model-based methods and deep learning-based methods. Image model-based methods are relatively mature, and there are a variety of algorithms. It can also be divided into two main categories: color balance model-based methods and optical model-based methods. The former mainly performs color correction on color images based on the characteristics of natural images, such as the grayscale world hypothesis. However, although different technologies have their own advantages, they also have some disadvantages:

1. It is quite difficult to obtain real-world underwater image datasets for deep learning. To solve this problem, synthetic image datasets are usually used for deep learning-based methods, but due to the complexity and diversity of underwater images, synthetic underwater images are far from real underwater images. Methods based on synthetic underwater images are effective for synthetic underwater images and certain types of real underwater images. However, these methods lack the ability to handle a wide variety of underwater images.

2. When detecting marine life, biologists need to obtain high-quality images or videos, and then manually annotate and identify species and their locations in these large-scale images and videos for further research. This is time-consuming, laborious and not suitable for real-time analysis on site.

Currently, deep convolutional neural networks (CNNs) have been widely used for more complex object detection in images and videos and have shown strong online detection capabilities. However, video surveillance in marine environments is very different from video surveillance in terrestrial environments, and deep CNNs for marine life detection face severe challenges:

1. Underwater images and videos suffer from strong absorption, scattering, color distortion, and noise from artificial light sources and marine snow particles, resulting in blurred, fuzzy images and blue or green tints.

2. It is often difficult for underwater objects to remain still, especially when marine life and underwater vehicles are also moving, resulting in motion blur and various postures of the creatures.

3. The underwater imaging environment is complex and changeable due to changes in lighting and spatiotemporal scales. Although low-level image enhancement can solve these problems to a certain extent and make underwater image data clearer, general image/video understanding algorithms still have difficulty handling such challenging problems in the marine environment.

In summary, the existing underwater target detection technology has certain defects when dealing with complex underwater environments.

Summary of the invention

In view of the above problems, the first aspect of the present invention proposes an underwater target detection model based on image enhancement and AR mechanism, including an image enhancement network module and a target detection module connected to each other;

The image enhancement network module adopts an end-to-end image enhancement network and designs a multi-scale loss function to perform underwater image enhancement training based on the collected image dataset Dataset to obtain enhanced underwater image network data; the image enhancement network module includes several FA modules, which are responsible for feature extraction and fusion. The FA modules treat different features and pixel areas equivalently, providing additional flexibility when processing different types of information;

The target detection module includes a feature extraction module and a detector module, and an AR mechanism is added thereto. The AR mechanism includes an improved Relation structure and an Attention structure. The Relation structure is used to improve the ability to capture channel relationships. The Attention structure adopts Self-Attention to capture the global features of the image and reduce the impact of background noise on the image. The target detection module is trained using an image data set enhanced by an image enhancement network module for underwater target detection.

Preferably, the image enhancement network module includes 6 FA modules, each FA module is composed of a channel attention module CA and a pixel attention module PA, CA can help the network better learn the importance of features between channels, and PA can enhance the interaction between pixels in the network.

Preferably, the channel attention module CA first uses a global average pool to bring the channel-wise global spatial information into the channel descriptor, as shown in the following formula:

Where g _c is the feature vector, F _c is the input feature, X _c (i, j) is the pixel value of the cth channel X _c at the pixel point with coordinates (i, j), H _p is the global pooling function, W is the row dimension of the image, and H is the column dimension of the image;

Then g _c is input into two convolutional layers, sigmoid activation function and ReLu activation function in sequence, as shown in the following formula:

CA _c =σ(Conv(δ(Conv(g _c ))))

Where CA _c is the weight of each channel, σ is the sigmoid activation function, δ is the ReLu activation function, and Conv is the convolutional layer;

Finally, perform a dot product operation on F _c and CA _c , as shown in the following formula:

in Represents the dot product operator.

Preferably, the pixel attention module PA first converts the feature vector output by CA into It is input into two convolutional layers, sigmoid activation function and ReLu activation function in sequence, as shown in the following formula:

PA＝σ(Conv(δ(Conv(F ^* ))))

Finally, perform a dot product operation on F ^* and PA, as shown in the following formula:

in is the feature vector output by PA.

Preferably, the design of a multi-scale loss function is specifically:

For the detected image I∈R ^C×H×W , C×H×W is the feature map of the shape of the I-th image, where W is the row dimension, H is the column dimension and C is the channel number dimension. The loss function of the end-to-end image enhancement network is constructed from three scales: reconstruction loss, perceptual loss and total variation loss, as shown in the following formula:

L _{＝ λrLr + λpLp + λTVLTV}

Among them, L _r , L _p , and L _TV correspond to the reconstruction loss, perceptual loss, and total variational loss of the image respectively; λ _r , λ _p , and λ _TV are the weight coefficients of the reconstruction loss, perceptual loss, and total variational loss;

The calculation formula of image reconstruction loss _Lr is:

in, is the reconstructed image I, ||·|| ₁ is the L ₁ norm;

The calculation formula of image perceptual loss _Lp is:

in, is the feature vector of image I, /> For image/> The eigenvector of , ||·|| ₂ is the L ₂ norm;

The calculation formula of the total variational loss L _TV is:

in, Indicates calculation/> The gradient along the W axis, /> Indicates calculation/> Gradient along the H axis.

Preferably, the improved Relation structure uses two fully connected layers to construct the improved relation structure to achieve modeling of the feature relationship between channels; two adjacent fully connected layers are used to model the relationship between channels, and the same number of channels as the input features are output, and the relation operation is defined as:

R(·)＝σ(W _c2 δ(W _c1 ))

Among them, W _c2 and W _c1 represent two fully connected layers, δ(·) represents the ReLU activation function, and σ(·) represents the Sigmoid activation function.

Preferably, the Attention structure adopts Self-Attention, and firstly calculates the query vector Q, key vector K and value vector V of the feature vector feat, as shown in the following formula:

Q＝W ^Q *feat

K＝W ^K *feat

V＝W ^K *feat

Where W ^Q , W ^K and W ^K are the weight parameter matrices of Q, K and V respectively;

Then calculate the attention weight a as shown below:

Where D _k is the modulus length of K;

Finally, the output vector is calculated:

Where _anj represents the weight of attention from the j-th input to the n-th output, _vj is the j-th element of V, and _hn is the attention vector.

A second aspect of the present invention provides an underwater target detection method based on image enhancement and AR mechanism, comprising the following process:

Acquire the target image to be detected;

Inputting the target image to be detected into the underwater target detection model as described in the first aspect;

Output target detection results.

The third aspect of the present invention provides an underwater target detection device based on image enhancement and AR mechanism, characterized in that: the device includes at least one processor and at least one memory, and the processor and memory are coupled; the memory stores a computer execution program of the underwater target detection model as described in the first aspect; when the processor executes the computer execution program stored in the memory, the processor executes an underwater target detection method based on image enhancement and AR mechanism.

The fourth aspect of the present invention provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program or instruction of the underwater target detection model as described in the first aspect, and when the program or instruction is executed by a processor, the processor executes an underwater target detection method based on image enhancement and AR mechanism.

Compared with the prior art, the present invention has the following beneficial effects:

(1) Adaptability: The end-to-end underwater image enhancement network still has strong adaptability in complex underwater environments and can adapt to different underwater scenes, which plays a positive role in underwater biometrics.

(2) Higher detection accuracy: The accuracy of underwater target detection is improved by introducing visual attention and relationship mechanisms.

(3) Real-time: This method does not require biologists to manually annotate and identify species and their locations from large-scale images and videos. It can be assembled and analyzed in real time.

In general, the method of this study improves the effect of underwater target detection by combining multiple factors such as adaptivity, visual attention, relational mechanism, and real-time, and is especially suitable for complex underwater scenes.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the technical solutions of the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, what is described below is only one embodiment of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

FIG1 is a diagram of an underwater target detection model based on the present invention combining the attention mechanism and the end-to-end image enhancement technology.

FIG. 2 is a schematic diagram of an enhanced network structure of the present invention.

FIG3 is a network framework diagram of the FA module of the present invention.

FIG. 4 is a schematic diagram of a target detection module in combination with an improved AR model according to the present invention.

FIG5 is a diagram showing the improved AR mechanism framework of the present invention.

FIG6 is a simplified structural block diagram of the underwater target detection device in Example 2.

Detailed ways

The invention will be further described below in conjunction with specific embodiments.

The organic combination of image enhancement technology and underwater target detection technology based on the attention mechanism and CNN network can effectively improve the accuracy of underwater target recognition and effectively enhance the ability to identify organisms in complex underwater environments. In the invention, an end-to-end underwater image enhancement network is first used, which combines multi-patch and multi-scale residual attention structures to enhance the diversity of real-world underwater images. The use of residual attention structures allows the network to select effective feature information. The method combines residual feature attention blocks and a novel combination of multi-scale and multi-block structures. The multi-block network extracts local features to adapt to various underwater images. It overcomes the problems of low contrast, unclear edges and color distortion in the original underwater images. Then, based on visual attention and relational mechanisms, a new method of applying an improved attention-relation (AR) module to an efficient marine organism detector (EMOD) is used for detecting marine organisms. This method can well improve the recognition ability of organisms in complex underwater environments.

The underwater target detection model process based on the combination of attention mechanism and end-to-end image enhancement technology of the present invention is shown in Figure 1, including a connected image enhancement network module and a target detection module;

The image enhancement network module adopts an end-to-end image enhancement network and designs a multi-scale loss function. It performs underwater image enhancement training based on the collected image dataset Dataset to obtain enhanced underwater image network data. The image enhancement network module contains several FA modules, which are responsible for feature extraction and fusion. The FA modules treat different features and pixel areas equivalently, providing additional flexibility when processing different types of information.

The target detection module includes a feature extraction module and a detector module, and an AR mechanism is added thereto. The AR mechanism includes an improved Relation structure and an Attention structure. The Relation structure is used to improve the ability to capture channel relationships, and the Attention structure adopts Self-Attention to capture the global features of the image and reduce the impact of background noise on the image. The target detection module is trained using an image data set enhanced by an image enhancement network module for underwater target detection.

1. Obtaining the image dataset

The present invention uses underwater cameras in different waters to monitor image data in real time. The collected data is an image of a unified standard size. The present invention sets the standard size of the image to 640*640 pixels. The quality of the image collected in each picture is different according to the actual environment. The video data of the underwater target to be detected is collected and the data processing results are integrated to obtain a data set Dataset. The data includes the order Perciformes, the order Cypriniformes and the order Scallopoides.

2. Image Enhancement Network

The end-to-end image enhancement network structure is shown in Figure 2. The network contains 6 FA modules, which are responsible for feature extraction and fusion. The FA modules treat different features and pixel regions equivalently, which can provide additional flexibility when processing different types of information. Each FA module consists of a channel attention module (CA) and a pixel attention module (PA). CA can help the network better learn the importance of features between channels, and PA can enhance the interaction between pixels in the network. The network framework of the FA module is shown in Figure 3.

The channel attention module CA first uses the global average pooling to bring the channel-wise global spatial information into the channel descriptor, as shown in the following formula:

Where g _c is the feature vector, F _c is the input feature, X _c (i, j) is the pixel value of the cth channel X _c at the pixel point with coordinates (i, j), H _p is the global pooling function, W is the row dimension of the image, and H is the column dimension of the image;

Then g _c is input into two convolutional layers, sigmoid activation function and ReLu activation function in sequence, as shown in the following formula:

CA _c =σ(Conv(δ(Conv(g _c ))))

Where CA _c is the weight of each channel, σ is the sigmoid activation function, δ is the ReLu activation function, and Conv is the convolutional layer;

Finally, perform a dot product operation on F _c and CA _c , as shown in the following formula:

in Represents the dot product operator.

The pixel attention module PA first converts the feature vector output by CA into It is input into two convolutional layers, sigmoid activation function and ReLu activation function in sequence, as shown in the following formula:

PA＝σ(Conv(δ(Conv(F ^* ))))

Finally, perform a dot product operation on F ^* and PA, as shown in the following formula:

in is the feature vector output by PA.

A multi-scale loss function is designed for an end-to-end image enhancement network. For the detected image I∈R ^C×H×W , C×H×W is the feature map of the shape of the I-th image, where W is the row dimension, H is the column dimension, and C is the channel number dimension. The loss function of the end-to-end image enhancement network is constructed from three scales: reconstruction loss, perceptual loss, and total variational loss, as shown in the following formula:

L _{＝ λrLr + λpLp + λTVLTV}

Among them, L _r , L _p , and L _TV correspond to the reconstruction loss, perceptual loss, and total variational loss of the image, respectively; λ _r , λ _p , and λ _TV are the weight coefficients of the reconstruction loss, perceptual loss, and total variational loss.

The calculation formula of image reconstruction loss _Lr is:

in, is the reconstructed image I, and ||·|| ₁ is the L ₁ norm.

The calculation formula of image perceptual loss _Lp is:

in, is the feature vector of image I, /> For image/> is the eigenvector of , and ||·|| ₂ is the L ₂ norm.

The calculation formula of the total variational loss L _TV is:

in, Indicates calculation/> The gradient along the W axis, /> Indicates calculation/> Gradient along the H axis.

3. Object Detection Module

The overall structure of the target detection module is shown in Figure 4, which includes a feature extraction module and a detector module, and an AR mechanism is added thereto. The AR mechanism includes an improved Relation structure and an Attention structure, as shown in Figure 5. The Relation structure is used to improve the ability to capture channel relationships, and the Attention structure uses Self-Attention to capture the global features of the image and reduce the impact of background noise on the image. The target detection module is trained using the image dataset enhanced by the image enhancement network module for underwater target detection.

The present invention improves the ability to capture channel relationships by improving the Relation structure in the AR module. The improved Relation structure uses two fully connected layers to construct the improved relation structure to achieve modeling of feature relationships between channels. Two adjacent fully connected layers are used to model the relationship between channels, and the same number of channels as the input features are output. The relation operation is defined as:

R(·)＝σ(W _c2 δ(W _c1 ))

Among them, W _c2 and W _c1 represent two fully connected layers, δ(·) represents the ReLU activation function, and σ(·) represents the Sigmoid activation function.

The improved Attention structure adopts Self-Attention. First, the query vector Q, key vector K and value vector V of the feature vector feat are calculated, as shown in the following formula:

Q＝W ^Q *feat

K＝W ^K *feat

V＝W ^K *feat

Where W ^Q , W ^K and W ^k are the weight parameter matrices of Q, K and V respectively;

Then calculate the attention weight a as shown below:

Where D _k is the modulus length of K;

Finally, the output vector is calculated:

Where _anj represents the weight of attention from the j-th input to the n-th output, _vj is the j-th element of V, and _hn is the attention vector.

4. Experimental results

In order to verify the underwater target detection method based on the combination of attention mechanism and end-to-end image enhancement technology proposed in the present invention, 700 carp, 700 perch and 700 scallop pictures were used as training sets, and 300 carp, 300 perch and 300 scallop pictures were used as test sets. The method proposed in the present invention is compared with the famous algorithms Faster R-CNN, YOLOv5 and SSD512. The evaluation index is the average precision (AP), which is calculated as the area under the precision-recall curve with IoU (intersection over union) thresholds from 0.5 to 0.95. Each algorithm is run five times independently, and the results are shown in Table 1.

Table 1 Experimental results

The method of the present invention has the highest average accuracy and the second lowest standard deviation, which shows that the method of the present invention has high classification accuracy for fish and good robustness.

Embodiment 2:

As shown in FIG6 , the present invention also provides an underwater target detection device based on image enhancement and AR mechanism, the device includes at least one processor and at least one memory, and also includes a communication interface and an internal bus; the memory stores a computer execution program of the underwater target detection model as described in Example 1; when the processor executes the computer execution program stored in the memory, the processor can execute an underwater target detection method based on image enhancement and AR mechanism. The internal bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application is not limited to only one bus or one type of bus. The memory may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk memory, and may also be a U disk, a mobile hard disk, a read-only memory, a disk or an optical disk.

The device may be provided as a terminal, a server or other forms of devices.

FIG6 is a block diagram of a device for exemplary purposes. The device may include one or more of the following components: a processing component, a memory, a power component, a multimedia component, an audio component, an input/output (I/O) interface, a sensor component, and a communication component. The processing component typically controls the overall operation of the electronic device, such as operations associated with display, phone calls, data communications, camera operations, and recording operations. The processing component may include one or more processors to execute instructions to complete all or part of the steps of the above method. In addition, the processing component may include one or more modules to facilitate interaction between the processing component and other components. For example, the processing component may include a multimedia module to facilitate interaction between the multimedia component and the processing component.

The memory is configured to store various types of data to support operations on the electronic device. Examples of such data include instructions for any application or method operating on the electronic device, contact data, phone book data, messages, pictures, videos, etc. The memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

The power supply assembly provides power to various components of the electronic device. The power supply assembly may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the electronic device. The multimedia assembly includes a screen that provides an output interface between the electronic device and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundaries of the touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia assembly includes a front camera and/or a rear camera. When the electronic device is in an operating mode, such as a shooting mode or a video mode, the front camera and/or the rear camera may receive external multimedia data. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.

The audio component is configured to output and/or input audio signals. For example, the audio component includes a microphone (MIC), and when the electronic device is in an operating mode, such as a call mode, a recording mode, and a voice recognition mode, the microphone is configured to receive an external audio signal. The received audio signal can be further stored in a memory or sent via a communication component. In some embodiments, the audio component also includes a speaker for outputting an audio signal. The I/O interface provides an interface between the processing component and the peripheral interface module, and the above-mentioned peripheral interface module can be a keyboard, a click wheel, a button, etc. These buttons may include, but are not limited to: a home button, a volume button, a start button, and a lock button.

The sensor assembly includes one or more sensors for providing various aspects of status assessment for the electronic device. For example, the sensor assembly can detect the on/off state of the electronic device, the relative positioning of components, such as the display and keypad of the electronic device, and the sensor assembly can also detect the position change of the electronic device or a component of the electronic device, the presence or absence of user contact with the electronic device, the orientation or acceleration/deceleration of the electronic device and the temperature change of the electronic device. The sensor assembly may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. The sensor assembly may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly may also include an accelerometer, a gyroscope sensor, a magnetic sensor, a pressure sensor or a temperature sensor.

The communication component is configured to facilitate wired or wireless communication between the electronic device and other devices. The electronic device can access a wireless network based on a communication standard, such as WiFi, 2G or 3G, or a combination thereof. In an exemplary embodiment, the communication component receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.

In an exemplary embodiment, the electronic device may be implemented by one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above methods.

Embodiment 3:

The present invention also provides a computer-readable storage medium, which stores a computer program or instruction of the underwater target detection model as described in Example 1. When the program or instruction is executed by a processor, the processor can execute an underwater target detection method based on image enhancement and AR mechanism.

Specifically, a system, device or equipment equipped with a readable storage medium may be provided, on which a software program code for implementing the functions of any of the above-mentioned embodiments is stored, and a computer or processor of the system, device or equipment reads and executes the instructions stored in the readable storage medium. In this case, the program code read from the readable medium itself can implement the functions of any of the above-mentioned embodiments, so the machine-readable code and the readable storage medium storing the machine-readable code constitute a part of the present invention.

The above storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk (such as CD-ROM, CD-R, CD-RW, DVD-20ROM, DVD-RAM, DVD-RW, DVD-RW), magnetic tape, etc. The storage medium can be any available medium that can be accessed by a general or special-purpose computer.

The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing/processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and/or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and/or edge servers. The network adapter card or network interface in each computing/processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing/processing device.

The computer program instructions for performing the operation of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be personalized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Although the above describes the specific implementation methods of the present invention, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims (10)

1. An underwater target detection model based on image enhancement and AR mechanism, characterized by: comprising an image enhancement network module and a target detection module connected; The image enhancement network module adopts an end-to-end image enhancement network and designs a multi-scale loss function to perform underwater image enhancement training based on the collected image dataset Dataset to obtain enhanced underwater image network data; the image enhancement network module includes several FA modules, which are responsible for feature extraction and fusion. The FA modules treat different features and pixel areas equivalently, providing additional flexibility when processing different types of information; The target detection module includes a feature extraction module and a detector module, and an AR mechanism is added thereto. The AR mechanism includes an improved Relation structure and an Attention structure. The Relation structure is used to improve the ability to capture channel relationships. The Attention structure adopts Self-Attention to capture the global features of the image and reduce the impact of background noise on the image. The target detection module is trained using an image data set enhanced by an image enhancement network module for underwater target detection. 2. An underwater target detection model based on image enhancement and AR mechanism as described in claim 1, characterized in that: the image enhancement network module includes 6 FA modules, each FA module is composed of a channel attention module CA and a pixel attention module PA, CA can help the network better learn the importance of features between channels, and PA can enhance the interaction between pixels in the network. 3. An underwater target detection model based on image enhancement and AR mechanism as claimed in claim 2, characterized in that: the channel attention module CA first uses a global average pool to bring the global spatial information of the channel into the channel descriptor, as shown in the following formula: Where g _c is the feature vector, F _c is the input feature, X _c (i, j) is the pixel value of the cth channel X _c at the pixel point with coordinates (i, j), H _p is the global pooling function, W is the row dimension of the image, and H is the column dimension of the image; Then g _c is input into two convolutional layers, sigmoid activation function and ReLu activation function in sequence, as shown in the following formula: CA _c =σ(Conv(δ(Conv(g _c )))) Where CA _c is the weight of each channel, σ is the sigmoid activation function, δ is the ReLu activation function, and Conv is the convolutional layer; Finally, perform a dot product operation on F _c and CA _c , as shown in the following formula: in Represents the dot product operator. 4. An underwater target detection model based on image enhancement and AR mechanism as claimed in claim 3, characterized in that: the pixel attention module PA first inputs the feature vector F _c ^* output by CA into two convolutional layers, sigmoid activation function and ReLu activation function in sequence, as shown in the following formula: Finally, perform a dot product operation on F ^* and PA, as shown in the following formula: in is the feature vector output by PA. 5. The underwater target detection model based on image enhancement and AR mechanism according to claim 1, characterized in that the multi-scale loss function is designed, specifically: For the detected image I∈R ^C×H×W , C×H×W is the feature map of the shape of the I-th image, where W is the row dimension, H is the column dimension and C is the channel number dimension. The loss function of the end-to-end image enhancement network is constructed from three scales: reconstruction loss, perceptual loss and total variation loss, as shown in the following formula: L _{＝ λrLr + λpLp + λTVLTV} Among them, L _r , L _p , and L _TV correspond to the reconstruction loss, perceptual loss, and total variational loss of the image respectively; λ _r , λ _p , and λ _TV are the weight coefficients of the reconstruction loss, perceptual loss, and total variational loss; The calculation formula of image reconstruction loss _Lr is: in, is the reconstructed image I, ||·|| ₁ is the L ₁ norm; The calculation formula of image perceptual loss _Lp is: in, is the feature vector of image I, /> For image/> The eigenvector of , ||·|| ₂ is the L ₂ norm; The calculation formula of the total variational loss L _TV is: in, Indicates calculation/> The gradient along the W axis, /> Indicates calculation/> Gradient along the H axis. 6. The underwater target detection model based on image enhancement and AR mechanism according to claim 1, characterized in that: the improved Relation structure uses two fully connected layers to construct the improved relation structure to achieve modeling of feature relationships between channels; two adjacent fully connected layers are used to model the relationship between channels, and the same number of channels as the input features are output, and the relation operation is defined as: R(·)＝σ(W _c2 δ(W _c1 )) Among them, W _c2 and W _c1 represent two fully connected layers, δ(·) represents the ReLU activation function, and σ(·) represents the Sigmoid activation function. 7. The underwater target detection model based on image enhancement and AR mechanism as claimed in claim 1, characterized in that: the Attention structure adopts Self-Attention, firstly calculates the query vector Q, key vector K and value vector V of the feature vector feat, as shown in the following formula: Q＝W ^Q *feat K＝W ^K *feat V＝W ^K *feat Where W ^Q , W ^K and W ^K are the weight parameter matrices of Q, K and V respectively; Then calculate the attention weight a as shown below: Where D _k is the modulus length of K; Finally, the output vector is calculated: Where _anj represents the weight of attention from the j-th input to the n-th output, _vj is the j-th element of V, and _hn is the attention vector. 8. An underwater target detection method based on image enhancement and AR mechanism, characterized by comprising the following processes: Acquire the target image to be detected; Inputting the target image to be detected into the underwater target detection model according to any one of claims 1 to 7; Output target detection results. 9. An underwater target detection device based on image enhancement and AR mechanism, characterized in that: the device includes at least one processor and at least one memory, the processor and the memory are coupled; the memory stores a computer execution program of the underwater target detection model as described in any one of claims 1 to 7; when the processor executes the computer execution program stored in the memory, the processor executes an underwater target detection method based on image enhancement and AR mechanism. 10. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program or instruction of an underwater target detection model as described in any one of claims 1 to 7, and when the program or instruction is executed by a processor, the processor executes an underwater target detection method based on image enhancement and AR mechanism.