Sea area enteromorpha identification method and system based on self-adaptive exposure adjustment

Abstract

According to the self-adaptive exposure adjustment sea area enteromorpha identification method and system provided by the invention, exposure adjustment is performed on an enteromorpha image through an improved M-Net network, the M-Net network integrates a SimSPPF module and an SE attention mechanism, the image quality is effectively enhanced while multi-scale detail information of the image is reserved, and the identification accuracy of the sea area enteromorpha is improved. The key features of the enteromorpha can be better captured under the condition of insufficient exposure or overexposure, recognition is performed based on the enteromorpha image after exposure adjustment, and the accuracy and stability of enteromorpha detection are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field related to enteromorpha image recognition, and in particular relates to a method and system for identifying enteromorpha in marine areas with adaptive exposure adjustment. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] With the continuous evolution of global climate change and marine ecological environment, the frequent occurrence and scale of marine disasters, especially green tide disasters (such as Enteromorpha outbreaks), have posed severe challenges to the ecological environment, economic activities and human life in coastal areas. The large-scale outbreak of Enteromorpha not only destroys the marine ecological balance and affects the quality of seawater, but also causes huge economic losses to marine fisheries, shipping and tourism. Therefore, real-time, accurate and efficient Enteromorpha monitoring and identification technology is particularly important in preventing and mitigating the impact of green tide disasters.

[0004] In the field of marine disaster monitoring and identification technology, real-time monitoring and accurate identification of Enteromorpha are of great significance for preventing marine ecological imbalance and protecting the marine environment and fishery resources. Traditional monitoring methods mostly rely on manual inspections or fixed monitoring stations. These methods are not only costly and inefficient, but also difficult to achieve large-scale, dynamic monitoring. The existing detection and identification algorithms based on remote sensing satellites and SAR are poor and have strict requirements on the observation environment. It is difficult to apply and identify in real time for a long time and over a large range. In recent years, with the rapid development of unmanned ship technology, it has become a new trend to monitor Enteromorpha in the sea area by using unmanned ships equipped with cameras.

[0005] However, the marine environment is complex and changeable, and the lighting conditions vary greatly. Traditional cameras are prone to exposure problems such as underexposure or overexposure in strong or weak light environments, which leads to a decrease in the quality of the Enteromorpha image and affects the recognition accuracy. Therefore, developing a method that can adaptively adjust exposure to cope with different lighting conditions to achieve high-precision recognition of Enteromorpha has become a technical problem that needs to be solved urgently in the current field of marine disaster monitoring. Summary of the invention

[0006] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a method and system for identifying Enteromorpha in marine areas with adaptive exposure adjustment. By adjusting the exposure of Enteromorpha images through an improved M-Net network, more accurate identification of Enteromorpha under different lighting conditions can be achieved.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] In a first aspect, the present invention provides a method for identifying marine enteromorpha with adaptive exposure adjustment, comprising:

[0009] Obtain under-exposed and over-exposed Enteromorpha images to construct a training set;

[0010] The improved M-Net network is trained using the constructed training set to obtain a trained exposure adjustment model;

[0011] Acquire an enteromorpha image to be identified, obtain an enteromorpha image with normal exposure based on a trained exposure adjustment model, and identify the enteromorpha image with normal exposure using the trained recognition model to obtain a recognition result;

[0012] Among them, the processing process of the improved M-Net network on the Enteromorpha image is as follows: based on each layer of the M-Net network, hierarchical features of different resolutions of the Enteromorpha image are extracted; for the hierarchical features extracted in each layer, the SimSPPF module is used to perform feature extraction and pooling operations to obtain a feature representation with multi-scale information, and the SE attention mechanism is used to dynamically adjust the importance of each channel of the feature representation, and the feature representation of the multi-scale information of each layer is cascaded with the hierarchical features of the next layer through pixel recombination, and the output results of each layer are fused and output through the CA attention module.

[0013] In a second aspect, the present invention provides a marine enteromorpha identification system with adaptive exposure adjustment, comprising:

[0014] A training set construction module is configured to: obtain under-exposed and over-exposed Enteromorpha images to construct a training set;

[0015] The model training module is configured as follows: using the constructed training set to train the improved M-Net network to obtain a trained exposure adjustment model; wherein the improved M-Net network processes the enteromorpha image as follows: extracting hierarchical features of different resolutions of the enteromorpha image based on each layer of the M-Net network; using the SimSPPF module to perform feature extraction and pooling operations on the hierarchical features extracted from each layer to obtain a feature representation with multi-scale information, using the SE attention mechanism to dynamically adjust the importance of each channel of the feature representation, and cascading the feature representation of the multi-scale information of each layer with the hierarchical features of the next layer through pixel reorganization, and fusing the output results of each layer through the CA attention module and then outputting;

[0016] The recognition module is configured to: obtain an enteromorpha image to be recognized, obtain an enteromorpha image with normal exposure based on a trained exposure adjustment model, and recognize the enteromorpha image with normal exposure using the trained recognition model to obtain a recognition result.

[0017] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in the first aspect is performed.

[0018] In a fourth aspect, the present invention provides a computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the method described in the first aspect is performed.

[0019] In a fifth aspect, the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements the method described in the first aspect.

[0020] One or more of the above technical solutions have the following beneficial effects:

[0021] In the present invention, the exposure of the Enteromorpha image is adjusted through an improved M-Net network. The M-Net network integrates the SimSPPF module and the SE attention mechanism, which effectively enhances the image quality while retaining the multi-scale detail information of the image. It can better capture the key features of the Enteromorpha under underexposure or overexposure conditions, and recognizes the Enteromorpha image based on the exposure-adjusted Enteromorpha image, thereby improving the accuracy and stability of Enteromorpha detection.

[0022] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0024] Figure 1 This is an overall flow chart of the method for identifying marine Enteromorpha in Embodiment 1 of the present invention;

[0025] Figure 2 Schematic diagram of the model structure of the MAIM-Net module in the first embodiment of the present invention;

[0026] Figure 3 This is a network structure diagram of the SimSPPF module in the first embodiment of the present invention;

[0027] Figure 4 This is a network structure diagram of the CA attention mechanism in Example 1 of the present invention;

[0028] Figure 5 Schematic diagram of the overall network model for Enteromorpha detection in Example 1 of the present invention. DETAILED DESCRIPTION

[0029] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0030] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0031] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0032] Embodiment 1

[0033] like Figure 1 As shown, this embodiment discloses a method for identifying marine Enteromorpha with adaptive exposure adjustment, comprising:

[0034] Step 1: Obtain under-exposed and over-exposed Enteromorpha images and construct a training set.

[0035] In step 1, a camera is installed on the unmanned boat to collect data about Enteromorpha and construct a data set. Specifically, LabelImg open source software is used to label Enteromorpha in the image, and the labeled content is the coordinates of the rectangular bounding box of Enteromorpha, which is stored for algorithm training and verification.

[0036] Step 2: Use the constructed training set to train the improved M-Net network to obtain a trained exposure adjustment model; wherein, the improved M-Net network processes the enteromorpha image as follows: extract hierarchical features of different resolutions of the enteromorpha image based on each layer of the M-Net network; use the SimSPPF module to perform feature extraction and pooling operations on the hierarchical features extracted from each layer to obtain a feature representation with multi-scale information, use the SE attention mechanism to dynamically adjust the importance of each channel of the feature representation, and concatenate the feature representation of the multi-scale information of each layer with the hierarchical features of the next layer through pixel reorganization, and fuse the output results of each layer before outputting them.

[0037] The improved M-Net network proposed in this embodiment is the MAIM-Net algorithm. Figure 2 As shown, the image is mainly adaptively adjusted for exposure, and an image with low exposure or overexposure is processed into an image with normal exposure, ensuring good image quality under different lighting conditions.

[0038] The MAIM-Net algorithm retains the spatial information of the image by adding SimSPPF to M-Net. The M-Net network integrates multi-attention mechanisms to effectively enhance the image quality while retaining multi-scale detail information, and better capture the key features of Enteromorpha under underexposure or overexposure conditions. By adjusting the exposure strength, a normally exposed image is formed, thereby improving the accuracy and stability of Enteromorpha detection.

[0039] The MAIM-Net algorithm uses bilinear UP(Down)Sample and Pixel Shuffle UP(Down)Sample as the main methods to improve the connection between resolution layers and increase the diversity of features. A 3×3 weight-shared convolution is used at each resolution of the input image, which is obtained by bilinear downsampling from the original resolution input. There are a total of 4 layers in the proposed MAIM-Net. The SimSPPF and SE attention mechanisms are combined into a SimSPPF-SE module and placed in each layer of MAIM-Net to extract sufficient and rich feature information. The Pixel Shuffle method is used as a downsampling module in the backbone feature path after the SimSPPF-SE module in the network to obtain more diverse and rich multi-scale cascade features. The feature map output by the SimSPPF-SE module after Pixel Shuffle is spliced ​​with the shallow features from bilinear downsampling, and similar operations are continued. Finally, the results of each layer are spliced ​​and sent to the CA attention module for feature fusion, and then aggregated with the input image to output the result.

[0040] MAIM-Net network Figure 2As shown, the network uses bilinear UP (Down) Sample and Pixel Shuffle UP (Down) Sample for up (down) sampling functions, and can better learn global information through layered construction. In this embodiment, the MAIM-Net network has a total of 4 layers. Specifically, the original resolution input image is bilinearly downsampled to obtain the input image of the corresponding encoder layer at each resolution, and the input image at each resolution is processed by 3×3 weight-sharing convolution. The feature map of the first layer of the encoder after weight-sharing convolution has two directions. One is to be directly spliced ​​to the corresponding upsampling end after being processed by the first SimSPPF-SE module; the other is to be processed by the first SimSPPF-SE module, and then pixel reorganized by the Pixel Shuffle method and spliced ​​with the feature map of the second layer corresponding to the weight-sharing convolution, and the spliced ​​feature map is processed by the second SimSPPF-SE module; the feature maps after weight-sharing convolution in the second to third layers have two directions, one is to be spliced ​​with the output of the first SimSPPF-SE module or the second SimSPPF-SE module after pixel reorganization of the previous layer, and the spliced ​​feature map is processed by the second SimSPPF-SE module or the third SimSPPF-SE module, and then directly spliced ​​to the corresponding upsampling end; the other is to be processed by the second SimSPPF-SE module or the third SimSPPF-SE module, and then spliced ​​by the Pixel Shuffle method. The Shuffle method is used to perform pixel reorganization and splice the feature map after weight-shared convolution processing of the next layer; the feature map of the fourth layer corresponding to the weight-shared convolution processing is spliced ​​with the output of the third SimSPPF-SE module after pixel reorganization of the third layer. The spliced ​​feature map is processed by the fourth SimSPPF-SE module and then directly spliced ​​to the corresponding upsampling end.

[0041] During upsampling, similarly, the output of the fourth SimSPPF-SE module corresponding to the fourth layer has two directions. One is to undergo convolution processing and bilinear sampling processing in sequence, and the other is to perform pixel reorganization through the Pixel Shuffle method and then splice it with the output of the third SimSPPF-SE module corresponding to the third layer. The spliced ​​features are then processed by the fifth SimSPPF-SE module, and the results of the processing by the fifth SimSPPF-SE module are sequentially processed by convolution and bilinear sampling; the first to third layers of the upsampling are also processed similarly; the features of each layer of the upsampling after convolution processing and bilinear sampling processing are spliced ​​and sent to the CA attention module.

[0042] SimSPPF module such as Figure 3As shown in the figure, the input feature map of the SimSPPF module is processed by the convolution layer, the BN layer and the ReLU activation function in turn to obtain the first extracted feature map; the first feature map is pooled by three maximum pooling layers in turn, the eigenvalues ​​are reduced to a fixed length, and the first pooled feature map corresponding to the output of each maximum pooling layer is obtained; the first pooled feature map of the fixed length obtained after each maximum pooling operation is merged with the first extracted feature map into an overall feature vector, and the overall feature vector is processed by the convolution layer, the BN layer and the ReLU activation function and then sent to the SE attention mechanism. The SE attention mechanism dynamically adjusts the importance of each channel by introducing the attention weights between channels, thereby improving the model's attention to different features, thereby improving the model's expression and generalization capabilities.

[0043] The SimSPPF module obtains feature representation with multi-scale information by performing feature extraction and pooling operations on image regions at different scales. It uses a lightweight pooling method to reduce the amount of calculation and improve model performance.

[0044] The CA attention mechanism in the MAIM-Net module is as follows Figure 4 As shown, it has high flexibility and can be easily integrated with various classic network structures to improve the network's feature extraction capabilities. The CA attention mechanism relies on channel information and divides the spliced ​​data into two one-dimensional feature codes, which encode each channel along the horizontal and vertical directions respectively. This method not only captures the dependency along one spatial direction, but also retains the precise position information along another spatial direction, enhancing the model's perception of different spatial positions and targets; average pooling is performed along different directions and then merged, and then processed by convolution, BN layer and Swish activation function in turn, and then convolution and Sigmoid function are performed respectively. The weighted feature map in the horizontal direction and the weighted feature map in the vertical direction are fused by element-by-element addition to form a new enhanced feature map. This enhanced feature map is added element-by-element to the original input feature map of the CA attention mechanism to retain the original feature information while fusing the attention information. The final output feature map is exactly the same size as the original input feature map.

[0045] After adaptive exposure adjustment through the MAIM-Net module, Enteromorpha detection is performed through YOLOv5. The overall network model of YOLOv5 is as follows: Figure 5As shown in the figure, the Backbone part of YOLOv5 is the backbone network in the algorithm, which is responsible for extracting the feature representation of the input image, and the Head part is mainly responsible for multi-scale target detection on the feature map extracted by the backbone network. The Backbone part of YOLOv5 is the backbone network in the algorithm, which is responsible for extracting the feature representation of the input image, and the Head part is mainly responsible for multi-scale target detection on the feature map extracted by the backbone network.

[0046] The Backbone network gradually increases the feature depth through convolutional layers. These convolutional layers include layers with different numbers of filters and different sizes of convolution kernels to gradually extract more complex features. The Head layer continues to process the backbone network output image and completes the pyramid pooling process through SSPF. First, the input feature map is pooled with different sizes to obtain a set of feature maps of different sizes. These feature maps are then connected together and reduced in dimension through a fully connected layer to finally obtain a feature vector of a fixed size, thereby improving the recognition ability of the neural network. The Head part adopts a multi-level feature fusion method. First, the feature map output by the backbone network is passed through a Conv module to reduce the number of channels and scale the feature map. Then, the feature maps of different levels are fused to obtain richer feature information, thereby improving the detection performance.

[0047] Specifically, the Backbone part gradually increases the feature depth through convolutional layers, which include layers with different numbers of filters and convolution kernels of different sizes to gradually extract more complex features. The CBS module integrates Conv layers for feature extraction, BN layers for normalization, and SiLU activation functions to enhance nonlinearity, working together to optimize the data processing flow. The C3 module is composed of three Conv layers with 3x3 convolution kernels. The stride of the first Conv layer is 2, which can halve the size of the feature map, and the stride of the second and third Conv layers is 1. Between each Conv layer, BN layers and SiLU activation functions are also added to improve the stability and generalization performance of the model. Resx consists of two CBS modules. By halving the size of the feature map, the network can pay more attention to the global information of the object, thereby improving the effect of feature extraction.

[0048] The head part continues to process the image output by the backbone network, and completes the pyramid pooling process through SSPF. First, the input feature map is pooled with different sizes to obtain a set of feature maps of different sizes. These feature maps are then connected together and reduced in dimension through a fully connected layer, and finally a feature vector of fixed size is obtained to improve the recognition ability of the neural network. The head part adopts a multi-level feature fusion method. First, the feature map output by the backbone network is reduced in dimension by a Conv module and the feature map is scaled. Then, the feature maps of different levels are fused to obtain richer feature information, thereby improving the detection performance.

[0049] Step 3: Obtain an enteromorpha image to be identified, obtain an enteromorpha image with normal exposure based on the trained exposure adjustment model, and identify the enteromorpha image with normal exposure using the trained recognition model to obtain a recognition result;

[0050] The present invention can be used for real-time monitoring of Enteromorpha by unmanned boats. The camera on the unmanned boat is used to capture images. The acquired images are firstly subjected to adaptive adjustment of image exposure by the MAIM-Net module, and then are recognized and detected by YOLOv5. The detection accuracy of Enteromorpha under different exposure levels can reach 95.6%.

[0051] The method proposed in this embodiment combines MAIM-Net with the YOLOv5 network to achieve accurate identification and monitoring of Enteromorpha under complex lighting conditions, effectively solving the problem of real-time monitoring of Enteromorpha in complex marine environments. The monitoring, identification, and tracking of Enteromorpha can be achieved by carrying a camera on an unmanned boat, and the movement of Enteromorpha can be monitored in real time, which improves the ability to warn of marine disasters and reduces the adverse effects of Enteromorpha on various industries in coastal areas.

[0052] The MAIM-Net algorithm of this embodiment mainly performs adaptive exposure adjustment of the image. The MAIM-Net algorithm retains the spatial information of the image by adding SimSPPF to the improved M-Net. The M-Net fusion multi-attention mechanism effectively enhances the image quality while retaining multi-scale detail information, and better captures the key features of the Enteromorpha under underexposure or overexposure conditions. Image loss is calculated by different loss functions, and the exposure strength is adjusted, thereby improving the accuracy and stability of Enteromorpha detection. After adaptive exposure adjustment through the MAIM-Net module, Enteromorpha detection is performed through YOLOv5.

[0053] A high-precision identification algorithm for marine Enteromorpha with adaptive exposure adjustment has been developed to solve the problem of identifying floating Enteromorpha algae in complex marine environments. It has been used to conduct detection at sea using unmanned ships equipped with cameras, achieving accurate identification of marine Enteromorpha, improving the ability to warn of marine disasters, and reducing the adverse effects of Enteromorpha on various industries in coastal areas.

[0054] Embodiment 2

[0055] The purpose of this embodiment is to provide a marine enteromorpha recognition system with adaptive exposure adjustment, comprising:

[0056] A training set construction module is configured to: obtain under-exposed and over-exposed Enteromorpha images to construct a training set;

[0057] The model training module is configured as follows: using the constructed training set to train the improved M-Net network to obtain a trained exposure adjustment model; wherein the improved M-Net network processes the enteromorpha image as follows: extracting hierarchical features of different resolutions of the enteromorpha image based on each layer of the M-Net network; using the SimSPPF module to perform feature extraction and pooling operations on the hierarchical features extracted from each layer to obtain a feature representation with multi-scale information, using the SE attention mechanism to dynamically adjust the importance of each channel of the feature representation, and cascading the feature representation of the multi-scale information of each layer with the hierarchical features of the next layer through pixel reorganization, and fusing the output results of each layer through the CA attention module and then outputting;

[0058] The recognition module is configured to: obtain an enteromorpha image to be recognized, obtain an enteromorpha image with normal exposure based on a trained exposure adjustment model, and recognize the enteromorpha image with normal exposure using the trained recognition model to obtain a recognition result.

[0059] In further embodiments, there is also provided:

[0060] An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in Embodiment 1 is performed. For the sake of brevity, no further description is given here.

[0061] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0062] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0063] A computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the method described in embodiment 1 is completed.

[0064] The method in the first embodiment can be directly embodied as a hardware processor, or a combination of hardware and software modules in the processor. The software module can be located in a mature storage medium in the field 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, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it will not be described in detail here.

[0065] A computer program product includes a computer program, and when the computer program is executed by a processor, the method described in the first embodiment is implemented.

[0066] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer executable instructions, such as instructions included in a program module, which are executed in a device on a real or virtual processor of the target to perform the process / method as described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or implement specific abstract data types. In various embodiments, the functions of program modules can be combined or divided between program modules as needed. Machine executable instructions for program modules can be executed in local or distributed devices. In distributed devices, program modules can be located in local and remote storage media.

[0067] The computer program code for implementing the method of the present invention can be written in one or more programming languages. These computer program codes can be provided to the processor of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the computer or other programmable data processing device, causes the function / operation specified in the flow chart and / or block diagram to be implemented. The program code can be executed completely on a computer, partially on a computer, as an independent software package, partially on a computer and partially on a remote computer or completely on a remote computer or server.

[0068] In the context of the present invention, computer program codes or related data may be carried by any appropriate carrier to enable a device, apparatus or processor to perform the various processes and operations described above. Examples of carriers include signals, computer readable media, etc. Examples of signals may include electrical, optical, radio, acoustic or other forms of propagation signals, such as carrier waves, infrared signals, etc.

[0069] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0070] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, 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

1. A method for identifying marine Enteromorpha with adaptive exposure adjustment, characterized in that: include: Obtain under-exposed and over-exposed Enteromorpha images to construct a training set; The improved M-Net network is trained using the constructed training set to obtain a trained exposure adjustment model; Acquire an enteromorpha image to be identified, obtain an enteromorpha image with normal exposure based on a trained exposure adjustment model, and identify the enteromorpha image with normal exposure using the trained recognition model to obtain a recognition result; The improved M-Net network processes the Enteromorpha image as follows: extracting hierarchical features of Enteromorpha images at different resolutions based on each layer of the M-Net network; For the hierarchical features extracted from each layer, the SimSPPF module is used to perform feature extraction and pooling operations to obtain a feature representation with multi-scale information. The SE attention mechanism is used to dynamically adjust the importance of each channel of the feature representation. The feature representation of the multi-scale information of each layer is cascaded with the hierarchical features of the next layer through pixel recombination, and the output results of each layer are fused and output.

2. The method for identifying marine enteromorpha with adaptive exposure adjustment according to claim 1, characterized in that: The coordinates of the rectangular bounding boxes of the Enteromorpha images under underexposure and overexposure are annotated, and a training set is constructed with the annotated Enteromorpha images. The improved M-Net network is trained using the constructed training set to obtain a trained exposure adjustment model.

3. The method for identifying marine enteromorpha with adaptive exposure adjustment according to claim 1, characterized in that: The SimSPPF module is used to perform feature extraction and pooling operations to obtain feature representation with multi-scale information, specifically: The input feature map is processed by a convolution layer, a normalization layer and an activation function in sequence to obtain a first extracted feature map; Processing the first extracted feature map through multiple maximum pooling layers in sequence, reducing the eigenvalues ​​to a fixed length, and obtaining a first pooling feature map corresponding to the output of each maximum pooling layer; Merge the first pooled feature map outputted by each maximum pooling layer and the first extracted feature map into an overall feature vector; The merged overall feature vector is processed in turn by the convolution layer, normalization layer and activation function to obtain a feature representation with multi-scale information.

4. The method for identifying marine enteromorpha with adaptive exposure adjustment according to claim 1, characterized in that: The output results of each layer are fused and output through the CA attention module, specifically: After splicing the output results of each layer, encode each channel of the spliced ​​data in the horizontal and vertical directions to obtain two one-dimensional encoding features; The two one-dimensional coding features obtained are merged after average pooling in different directions, and the merged features are processed by convolution layer, normalization layer and activation function in sequence to obtain a second extracted feature map; After convolution and activation function processing of the second extracted feature map, a horizontal weighted feature map and a vertical weighted feature map are obtained; The horizontal weighted feature map and the vertical weighted feature map are merged to obtain an enhanced feature map; The enhanced feature map and the original input features of the CA attention module are added element by element to obtain the output of the CA attention module.

5. The method for identifying marine enteromorpha with adaptive exposure adjustment according to claim 1, characterized in that: The improved M-Net network's processing process for Enteromorpha images is as follows: Perform bilinear downsampling on the original resolution input image to obtain the input image of the corresponding encoder layer at each resolution; The input image at each resolution is processed by weight-shared convolution respectively. The feature maps obtained by weight-shared convolution processing of each layer of the encoder are processed by SimSPPF-SE modules and directly spliced ​​to the corresponding upsampling end. The pixels are reorganized by Pixel Shuffle method and spliced ​​with the feature maps of the next layer processed by weight-shared convolution. The spliced ​​feature maps are processed by the corresponding SimSPPF-SE modules and spliced ​​to the corresponding upsampling end. During upsampling, the output of each layer of SimSPPF-SE module is processed by convolution and bilinear sampling, and the pixel is reorganized by the Pixel Shuffle method and then spliced ​​with the output of the corresponding SimSPPF-SE module of the previous layer. The spliced ​​features are processed by the corresponding SimSPPF-SE module and then convolution and bilinear sampling are performed. The features of each upsampling layer after convolution and bilinear sampling are concatenated to obtain the output features.

6. The method for identifying marine enteromorpha with adaptive exposure adjustment according to claim 1, characterized in that: The YOLOv5 network is used as the recognition model.

7. A marine enteromorpha recognition system with adaptive exposure adjustment, characterized in that: include: A training set construction module is configured to: obtain under-exposed and over-exposed Enteromorpha images to construct a training set; The model training module is configured as follows: using the constructed training set to train the improved M-Net network to obtain a trained exposure adjustment model; wherein the improved M-Net network processes the enteromorpha image as follows: extracting hierarchical features of different resolutions of the enteromorpha image based on each layer of the M-Net network; using the SimSPPF module to perform feature extraction and pooling operations on the hierarchical features extracted from each layer to obtain a feature representation with multi-scale information, using the SE attention mechanism to dynamically adjust the importance of each channel of the feature representation, and cascading the feature representation of the multi-scale information of each layer with the hierarchical features of the next layer through pixel reorganization, and fusing the output results of each layer through the CA attention module and then outputting; The recognition module is configured to: obtain an enteromorpha image to be recognized, obtain an enteromorpha image with normal exposure based on a trained exposure adjustment model, and recognize the enteromorpha image with normal exposure using the trained recognition model to obtain a recognition result.

8. An electronic device, characterized in that: The method comprises a memory and a processor and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method according to any one of claims 1 to 6 is completed.

9. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the method described in any one of claims 1 to 6.

10. A computer program product, characterized in that The invention comprises a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 6.

Images