An image enhancement processing method for fishery resources statistics

Abstract

The application provides an image enhancement processing method for fishery resource statistics, belongs to the field of image enhancement, and aims to build a color prior feature enhancement network, the color prior feature enhancement network comprises a high-frequency information enhancement module and a low-frequency information enhancement module, the high-frequency information enhancement module strengthens the detail features in the fishery resource statistics image, the low-frequency information enhancement module focuses on the accurate correction of the image color, better retains the overall content and color expression of the fishery resource statistics image, a frequency information interaction module is built, interaction between the high-frequency information and the low-frequency information is realized, the color information in the low-frequency information is effectively fused with the texture and detail features in the high-frequency information, and fishery resource statistics image enhancement is realized.

Description

Technical Field

[0001] This invention belongs to the field of image enhancement, and specifically relates to an image enhancement processing method for fishery resource statistics. Background Technology

[0002] In fisheries resource management, accurately obtaining information on the quantity, distribution, and behavioral characteristics of fish populations is crucial. Due to the complexity of the underwater environment, the quality of fisheries resource statistical images is greatly affected, resulting in problems such as low contrast, high noise, and color distortion, which in turn affects the accuracy of fish identification and statistics. Therefore, how to effectively enhance the images acquired by underwater camera equipment has become a key technology to improve the accuracy of fisheries resource statistics. Image enhancement techniques mainly include noise reduction, contrast enhancement, color correction, and edge detection. By removing the scattered light effect and uneven illumination in the underwater environment, the details and color information of the image can be restored, and the visibility of fish can be enhanced. In this way, the accuracy of subsequent fish identification, classification, and quantity statistics will be greatly improved, providing a reliable data foundation for the accurate assessment of fisheries resources.

[0003] In recent years, diffusion-based methods have attracted widespread attention for their outstanding performance in image restoration tasks. Although standard diffusion models have demonstrated sufficient capabilities, unpredictable artifacts may occur due to the diversity introduced during the sampling process from randomly generated Gaussian noise. Furthermore, diffusion models have limited ability to focus on fine-grained information, leading to the neglect of texture and details. Most previous methods are based on the original pixel space of the image, with limited exploration of the frequency space characteristics of underwater images. This results in the inability to effectively utilize the representational capabilities of depth models to generate high-quality images. To address these issues, this invention proposes an image enhancement processing method for fishery resource statistics. A feature information enhancement module is constructed to fully explore the frequency domain information in fishery resource statistics images. This achieves detail enhancement of high-frequency information and color correction of low-frequency information. Simultaneously, the interaction between high-frequency and low-frequency information is realized, further enhancing image quality. Summary of the Invention

[0004] This invention provides an image enhancement processing method for fishery resource statistics. The method aims to construct a color prior feature enhancement network, use discrete wavelet transform to divide the image into high-frequency information containing global structure and texture details and low-frequency information containing content and color, enhance the frequency information in the wavelet space, separate the high-frequency and low-frequency information for detail enhancement and color correction respectively, and construct a frequency information interaction module to realize the interaction between high-frequency and low-frequency information, thereby achieving image enhancement.

[0005] This invention aims to propose a color prior feature enhancement network and provide an image enhancement processing method for fishery resource statistics, including the following steps.

[0006] S1. Collection of fishery resource statistical images: An underwater camera is installed on an underwater robot to cruise and take pictures in a specific water area to obtain underwater fishery resource statistical images. The obtained fishery resource statistical images are labeled and compiled into a dataset.

[0007] S2. Construct a color prior feature enhancement network, which consists of N feature information enhancement modules, including a low-frequency information enhancement module, a high-frequency information enhancement module, and a frequency information interaction module.

[0008] S3. Construct a low-frequency information enhancement module, including SE attention mechanism, regularization, GELU activation function, fully connected layer and multi-scale color information extraction module, to realize low-frequency information color correction of fishery resource statistical images.

[0009] S4. Construct a high-frequency information enhancement module, including a 1×1 convolutional layer, a deep convolutional layer, an attention mechanism, and a multilayer perceptron, to enhance the high-frequency information details of fishery resource statistical images.

[0010] S5. Construct a frequency information interaction module, including a 1×1 convolutional layer, a 3×3 depth convolutional layer, and a softmax activation function, to achieve mutual reinforcement of high and low frequency information.

[0011] S6. Fishery resource statistical image processing: Input the fishery resource statistical image to be processed into the color prior feature enhancement network to obtain the enhanced fishery resource statistical image.

[0012] Preferably, in step S2, for the color prior feature enhancement network, the input is a fishery resource statistical image. H, W and C represent respectively The height, width, and channels are determined using discrete wavelet transform. Decomposed into high-frequency information and low-frequency information, , Represents discrete wavelet transform. This represents the low-frequency information of the input fishery resource statistics image, including the image's content and color information. This inputs high-frequency information from a fisheries resource statistical image, including detailed information on global structure and texture. Then, the high-frequency and low-frequency information are input into a feature enhancement module, which consists of a low-frequency information enhancement module, a high-frequency information enhancement module, and a frequency information interaction module. The low-frequency information... Obtained through the low-frequency information enhancement module High-frequency information After being obtained through the high-frequency information enhancement module , and The input to the frequency information interaction module enables information exchange between high-frequency and low-frequency information. , This indicates the frequency information interaction module. This indicates low-frequency information following the interaction. Indicate the high-frequency information after the interaction, and then and Adding elements together yields , This indicates the addition of elements. and Adding elements together yields , This indicates the addition of elements. and As input to the next feature information enhancement module, it is obtained after passing through N feature information enhancement modules. and , and The final image is generated after inverse discrete wavelet transform. , This represents the inverse discrete wavelet transform.

[0013] Preferably, in step S2, the feature information enhancement module performs targeted enhancement on low-frequency and high-frequency information respectively, and realizes the interaction between high-frequency and low-frequency information through the frequency information interaction module. Information in different frequency bands is processed independently to improve the overall visual quality of the image. At the same time, through multiple frequency interactions and feature enhancements, the features of the image are gradually optimized. Combined with discrete wavelet inverse transform, the optimized high-frequency and low-frequency information is reconstructed into the enhanced image, effectively maintaining the global structural consistency and detail clarity of the image, and realizing detail enhancement and color correction.

[0014] Preferably, in step S3, for the low-frequency information enhancement module, the low-frequency information of the fishery resource statistical image is input. H, W, and C represent height, width, and channel, respectively. First, the SE attention mechanism adaptively assigns weights to different feature channels, enhancing important features and suppressing unimportant ones. Dropout is used for regularization to prevent overfitting during model training. A 3×3 convolutional layer and the GELU activation function work together to capture low-frequency information while improving the model's representational power, resulting in optimized low-frequency information. , This indicates the SE attention mechanism. This indicates Dropout regularization. This represents a 3×3 convolutional layer. The GELU activation function is used, and then a multi-scale color information extraction module is used to obtain more comprehensive color information. , This indicates the multi-scale color information extraction module, which will ultimately... The input is fed into a fully connected layer to obtain enhanced low-frequency information. .

[0015] Preferably, in step S3, the low-frequency information enhancement module optimizes the low-frequency information of the fishery resource statistics image through multiple steps. First, the SE attention mechanism is used to adaptively weight the feature channels, which can enhance important features and suppress irrelevant features, ensuring that the model focuses on information that plays a key role in fishery resource statistics. Regularization is performed through Dropout to effectively prevent overfitting during training. Combining 3×3 convolutional layers and the GELU activation function, it can not only capture local features in low-frequency information, but also improve the model's nonlinear representation ability and detail capture ability. The multi-scale color information extraction module further obtains richer color information from different scales, thereby providing the model with more comprehensive and accurate color features. Finally, the optimized low-frequency information is output through a fully connected layer, ensuring that the low-frequency information not only retains the overall content of the image, but also further enhances the color processing.

[0016] Preferably, in step S3, for the multi-scale color information extraction module, in order to process and extract the information of each color channel separately, the features of the input fishery resource statistical image are first processed. Extract the R, G, and B color channels separately. , This approach separates color channels, using 3×3 convolutional layers and the ReLU activation function to extract local features from each channel. Then, a SE attention mechanism is applied to assign different weights to the R, G, and B color channels, enhancing important features and suppressing less important ones. The color features processed by the channel attention mechanism are then fused element-wise. Finally, layer normalization is performed on the fused color features to obtain multi-scale color information. , This represents a 3×3 convolutional layer. Represents the ReLU activation function. This indicates the SE attention mechanism. Presentation layer normalization processing.

[0017] Preferably, in step S3, the multi-scale color information extraction module extracts the R, G, and B color channels separately, enabling independent processing of information for each color channel. This separate processing method allows the model to deeply explore the local features of each channel. By using 3×3 convolutional layers and the ReLU activation function to enhance feature representation, and introducing the SE attention mechanism to assign weights to different color channels, the model effectively increases the attention to important features and suppresses unimportant features, thereby optimizing the efficiency and accuracy of information extraction. Finally, by element-wise addition and fusion of the features of each channel and layer normalization, the balance and consistency of color information are ensured. This design not only improves the representation ability of color features but also enhances the adaptability and performance of the model in complex image processing, making it more effective and reliable in the analysis of fishery resource statistical images.

[0018] Preferably, in step S4, for the high-frequency information enhancement module, the high-frequency information of the fishery resource statistical image is first enhanced. Perform layer normalization processing H, W, and C represent height, width, and channel, respectively. The representation layers are normalized, and the processed features are then passed through 1×1 convolutional layers and 7×7 depthwise convolutional layers, 1×1 convolutional layers and 5×5 depthwise convolutional layers, and 1×1 convolutional layers and 3×3 depthwise convolutional layers, respectively. , , , These represent the query, key, and value, respectively. This represents a 1×1 convolutional layer. This represents a 7×7 depth convolutional layer. This represents a 5×5 depth convolutional layer. This represents a 3×3 deep convolutional layer. Self-attention calculation, layer normalization, and multi-layer perceptron operations are then performed on Q, K, and V. Weighted values ​​of features are calculated to enhance the model's focus on important features. Finally, residual connections are introduced to improve the stability of model training. , This indicates the self-attention mechanism. Presentation layer normalization processing, This represents a multi-layer sensing operation. This indicates the addition of elements.

[0019] Preferably, in step S4, the high-frequency information enhancement module effectively improves the processing capability of high-frequency information in fishery resource statistical images. Through 1×1 convolutional layers and deep convolutional layers with different kernels, the module can capture multi-scale detailed features from different receptive fields and generate query, key, and value features respectively. Then, the self-attention mechanism can focus on important details in high-frequency information by calculating the correlation of Q, K, and V, and improve the model's ability to identify key features. Combined with a multilayer perceptron, the feature information is further processed. Finally, residual connections are used to add the input high-frequency information with the processed features, effectively avoiding the gradient vanishing problem while maintaining the continuous flow of information. This not only enhances the detail representation of the image but also improves the training stability of the model.

[0020] Preferably, in step S5, the frequency information interaction module functions to facilitate information interaction between low-frequency and high-frequency information, achieving mutual reinforcement of high and low-frequency information, and inputting high-frequency information. and low-frequency information H, W, and C represent height, width, and channel, respectively, for high-frequency information. After a 1×1 convolutional layer, we obtain , This represents a 1×1 convolutional layer. express Query, high-frequency information After passing through a 1×1 convolutional layer and a 3×3 depthwise convolutional layer, we obtain... , This represents a 1×1 convolutional layer. This represents a 3×3 depth convolutional layer. express Values, low-frequency information After a 1×1 convolutional layer, we obtain , This represents a 1×1 convolutional layer. express The key, low-frequency information After passing through a 1×1 convolutional layer and a 3×3 depthwise convolutional layer, we obtain... , This represents a 1×1 convolutional layer. This represents a 3×3 depth convolutional layer. express The value, and then for and Softmax normalization is performed to obtain , This represents the softmax activation function, and finally... and Element-wise multiplication yields high-frequency information after interaction. , Indicates element-wise multiplication, and Element-wise multiplication yields low-frequency information after interaction. , This indicates element-wise multiplication.

[0021] Preferably, in step S5, the frequency information interaction module promotes the interaction between low-frequency and high-frequency information to achieve mutual reinforcement of high and low frequency information. This module first extracts the query of high-frequency information and the key of low-frequency information through a 1×1 convolutional layer, obtains their respective values ​​using a deep convolutional layer, and then processes the query and key through a softmax normalization operation to generate a weight matrix. This matrix captures the relationship between high-frequency and low-frequency information. Finally, the weight matrix is ​​multiplied by the high-frequency and low-frequency values ​​to achieve targeted reinforcement of information. This design not only improves the accuracy and effectiveness of feature extraction, but also enhances the model's adaptability when processing complex images, ensuring the combination of high-frequency details and low-frequency information.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0023] The color prior feature enhancement network provided by this invention uses discrete wavelet transform to divide an image into high-frequency information containing global structure and texture details and low-frequency information containing content and color. Frequency information enhancement is performed in wavelet space, and high-frequency and low-frequency information are processed separately to handle detail enhancement and color correction problems respectively. At the same time, a frequency information interaction module is constructed to realize the interaction between high-frequency and low-frequency information, thereby enhancing image quality. Attached Figure Description

[0024] Figure 1 This is a flowchart of an image enhancement processing method for fishery resource statistics provided by the present invention;

[0025] Figure 2 This is a diagram of the color prior feature enhancement network structure provided by the present invention;

[0026] Figure 3 This is a structural diagram of the low-frequency information enhancement module provided by the present invention;

[0027] Figure 4 This is a structural diagram of the high-frequency information enhancement module provided by the present invention;

[0028] Figure 5 This is a comparison image of the fishery resource statistical image before and after enhancement provided by the present invention. Detailed Implementation

[0029] This invention provides an image enhancement processing method for fishery resource statistics. It aims to construct a color prior feature enhancement network, which includes a high-frequency information enhancement module and a low-frequency information enhancement module. The high-frequency information enhancement module strengthens edges, details, and complex textures in the image, making the detailed features of the fishery resource statistics image clearer. The low-frequency information enhancement module focuses on accurate color correction, ensuring consistency and accuracy of image color information during the enhancement process, thereby better preserving the overall content and color expression of the fishery resource statistics image. A frequency information interaction module is also constructed to realize the interaction between high-frequency and low-frequency information. The color and global content information in the low-frequency information can be effectively integrated with the texture and detail features in the high-frequency information, thus achieving fishery resource statistics image enhancement.

[0030] Please see Figure 1 As shown in the embodiment of this application, there is an image enhancement processing method for fishery resource statistics.

[0031] S1. Collection of fishery resource statistical images: An underwater camera is installed on an underwater robot to cruise and take pictures in a specific water area to obtain 2,000 underwater fishery resource statistical images. The collected fishery resource statistical images are labeled, and the labeled images constitute a fishery resource statistical image dataset. The 2,000 fishery resource statistical image dataset is divided into a training set of 1,400 images and a validation set of 600 images.

[0032] S2. Construct a color prior feature enhancement network, consisting of 20 feature information enhancement modules, including a low-frequency information enhancement module, a high-frequency information enhancement module, and a frequency information interaction module.

[0033] Furthermore, such as Figure 2 As shown, input a statistical image of fishery resources. 640, 640 and 3 respectively represent The height, width, and channels are determined using discrete wavelet transform. Decomposed into high-frequency information and low-frequency information, , Represents discrete wavelet transform. This represents the low-frequency information of the input fishery resource statistics image, including the image's content and color information. This inputs high-frequency information from a fisheries resource statistical image, including detailed information on global structure and texture. Then, the high-frequency and low-frequency information are input into a feature enhancement module, which consists of a low-frequency information enhancement module, a high-frequency information enhancement module, and a frequency information interaction module. The low-frequency information... After being obtained by the low-frequency information enhancement module High-frequency information After being obtained through the high-frequency information enhancement module , and The input to the frequency information interaction module enables information exchange between high-frequency and low-frequency information. , This indicates the frequency information interaction module. This indicates low-frequency information following the interaction. Indicate the high-frequency information after the interaction, and then and Adding elements together yields , This indicates the addition of elements. and Adding elements together yields , This indicates the addition of elements. and As input to the next feature enhancement module, after passing through 20 feature enhancement modules, the result is... and , and The final image is generated after inverse discrete wavelet transform. , This represents the inverse discrete wavelet transform.

[0034] S3. Construct a low-frequency information enhancement module, including SE attention mechanism, regularization, GELU activation function, fully connected layer and multi-scale color information extraction module, to realize low-frequency information color correction of fishery resource statistical images.

[0035] Furthermore, such as Figure 3 As shown, for the low-frequency information enhancement module, the input is the low-frequency information of the fishery resource statistical image. 320, 320, and 3 represent height, width, and channel, respectively. First, the SE attention mechanism adaptively assigns weights to different feature channels, enhancing important features and suppressing unimportant ones. Dropout is used for regularization to prevent overfitting during model training. A 3×3 convolutional layer and the GELU activation function work together to capture low-frequency information while improving the model's representational power, resulting in optimized low-frequency information. , This indicates the SE attention mechanism. This indicates Dropout regularization. This represents a 3×3 convolutional layer. The GELU activation function is used, and then a multi-scale color information extraction module is used to obtain more comprehensive color information. , This indicates the multi-scale color information extraction module, which will ultimately... The input is fed into a fully connected layer to obtain enhanced low-frequency information. .

[0036] S4. Construct a high-frequency information enhancement module, including a 1×1 convolutional layer, a deep convolutional layer, an attention mechanism, and a multilayer perceptron, to enhance the high-frequency information details of fishery resource statistical images.

[0037] Furthermore, such as Figure 4 As shown, for the high-frequency information enhancement module, the high-frequency information of the fishery resource statistical image is first enhanced. Perform layer normalization processing 320, 320 and 3 represent the height, width and channel, respectively. The representation layer is normalized, and the processed features are then passed through 1×1 convolutional layers and 7×7 depthwise convolutional layers, 1×1 convolutional layers and 5×5 depthwise convolutional layers, and 1×1 convolutional layers and 3×3 depthwise convolutional layers, respectively. , , , These represent the query, key, and value, respectively. This represents a 1×1 convolutional layer. This represents a 7×7 depth convolutional layer. This represents a 5×5 depth convolutional layer. This represents a 3×3 deep convolutional layer. Self-attention calculation, layer normalization, and multi-layer perceptron operations are then performed on Q, K, and V. Weighted values ​​of features are calculated to enhance the model's focus on important features. Finally, residual connections are introduced to improve the stability of model training. , This indicates the self-attention mechanism. Presentation layer normalization processing, This represents a multi-layer sensing operation. This indicates the addition of elements.

[0038] S5. Construct a frequency information interaction module, including a 1×1 convolutional layer, a 3×3 depth convolutional layer, and a softmax activation function, to achieve mutual reinforcement of high and low frequency information.

[0039] Furthermore, the frequency information interaction module facilitates information exchange between low-frequency and high-frequency information, enabling mutual reinforcement of high and low-frequency information, and inputting high-frequency information. and low-frequency information 320, 320 and 3 respectively represent and High, width and channels, high frequency information After a 1×1 convolutional layer, we obtain , This represents a 1×1 convolutional layer. express Query, high-frequency information After passing through a 1×1 convolutional layer and a 3×3 depthwise convolutional layer, we obtain... , This represents a 1×1 convolutional layer. This represents a 3×3 depth convolutional layer. express Values, low-frequency information After a 1×1 convolutional layer, we obtain , This represents a 1×1 convolutional layer. express The key, low-frequency information After passing through a 1×1 convolutional layer and a 3×3 depthwise convolutional layer, we obtain... , This represents a 1×1 convolutional layer. This represents a 3×3 depth convolutional layer. express The value, and then for and Softmax normalization is performed to obtain , This represents the softmax activation function, and finally... and Element-wise multiplication yields high-frequency information after interaction. , Indicates element-wise multiplication, and Element-wise multiplication yields low-frequency information after interaction. , This indicates element-wise multiplication.

[0040] S6. Fishery resource statistical image processing: Input the fishery resource statistical image to be processed into the color prior feature enhancement network to obtain the enhanced fishery resource statistical image.

[0041] Furthermore, in step S6, the color prior feature enhancement network is developed using the Python language through the PyCharm application and implemented based on the PyTorch framework. The color prior feature enhancement network takes a fishery resource statistical image to be processed with a resolution of 640×640×3 as input, and starts training using a self-supervised learning method to effectively extract high-frequency and low-frequency information in the image and realize the interaction between high-frequency and low-frequency information. Finally, it outputs the enhanced fishery resource statistical image.

[0042] Furthermore, such as Figure 5 As shown, Figure 5 The left-middle half is a statistical image of fishery resources to be processed. Figure 5The right half of the image shows the enhanced fishery resource statistics image after processing by a color prior feature enhancement network. This significantly improves the clarity and recognizability of the fishery resource statistics image, thereby enhancing the accuracy of subsequent fish identification, classification, and quantity statistics.

[0043] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept of the present invention, and these modifications and improvements all fall within the protection scope of the present invention.

Claims

1. An image enhancement processing method for fishery resource statistics, characterized in that, Includes the following steps: S1. Collection of fishery resource statistical images: An underwater camera is installed on an underwater robot to cruise and take pictures in a specific water area to obtain underwater fishery resource statistical images. The obtained fishery resource statistical images are manually labeled and compiled into a dataset. S2. Construct a color prior feature enhancement network, consisting of N feature information enhancement modules. These modules include a low-frequency information enhancement module, a high-frequency information enhancement module, and a frequency information interaction module. The low-frequency information enhancement module extracts multi-scale color information. To process and extract information from each color channel separately, the features of the input fishery resource statistical image are first analyzed. Extract the R, G, and B color channels separately. , This approach separates color channels, using 3×3 convolutional layers and the ReLU activation function to extract local features from each channel. Then, a SE attention mechanism is applied to assign different weights to the R, G, and B color channels, enhancing important features and suppressing less important ones. The color features processed by the channel attention mechanism are then fused element-wise. Finally, layer normalization is performed on the fused color features to obtain multi-scale color information. , This represents a 3×3 convolutional layer. Represents the ReLU activation function. This indicates the SE attention mechanism. Presentation layer normalization processing; S3. Construct a low-frequency information enhancement module, including an SE attention mechanism, regularization, GELU activation function, fully connected layers, and a multi-scale color information extraction module, to achieve low-frequency color correction of fishery resource statistical images. For the low-frequency information enhancement module, the low-frequency information of the fishery resource statistical image is input. H, W, and C represent height, width, and channel, respectively. First, the SE attention mechanism adaptively assigns weights to different feature channels, enhancing important features and suppressing unimportant ones. Dropout is used for regularization to prevent overfitting during model training. A 3×3 convolutional layer and the GELU activation function work together to capture low-frequency information while improving the model's representational power, resulting in optimized low-frequency information. , This indicates the SE attention mechanism. This indicates Dropout regularization. This represents a 3×3 convolutional layer. The GELU activation function is used, and then a multi-scale color information extraction module is used to obtain more comprehensive color information. , This indicates the multi-scale color information extraction module, which will ultimately... The input is fed into a fully connected layer to obtain enhanced low-frequency information. ; S4. Construct a high-frequency information enhancement module, including a 1×1 convolutional layer, a deep convolutional layer, an attention mechanism, and a multilayer perceptron, to enhance the high-frequency information details of the fishery resource statistical image. For the high-frequency information enhancement module, firstly, the high-frequency information of the fishery resource statistical image... Perform layer normalization processing H, W, and C represent height, width, and channel, respectively. The representation layers are normalized, and the processed features are then passed through 1×1 convolutional layers and 7×7 depthwise convolutional layers, 1×1 convolutional layers and 5×5 depthwise convolutional layers, and 1×1 convolutional layers and 3×3 depthwise convolutional layers, respectively. , , , These represent the query, key, and value, respectively. This represents a 1×1 convolutional layer. This represents a 7×7 depth convolutional layer. This represents a 5×5 depth convolutional layer. This represents a 3×3 deep convolutional layer. Self-attention calculation, layer normalization, and multi-layer perceptron operations are then performed on Q, K, and V. Weighted values ​​of features are calculated to enhance the model's focus on important features. Finally, residual connections are introduced to improve the stability of model training. , This indicates the self-attention mechanism. Presentation layer normalization processing, This represents a multi-layer sensing operation. This indicates element addition; S5. Construct a frequency information interaction module, including a 1×1 convolutional layer, a 3×3 depthwise convolutional layer, and a softmax activation function, to achieve mutual reinforcement of high and low frequency information. The function of this module is to promote information interaction between low and high frequency information, achieving mutual reinforcement of high and low frequency information, and inputting high-frequency information. and low-frequency information H, W, and C represent height, width, and channel, respectively, for high-frequency information. After a 1×1 convolutional layer, we obtain , This represents a 1×1 convolutional layer. express Query, high-frequency information After passing through a 1×1 convolutional layer and a 3×3 depthwise convolutional layer, we obtain... , This represents a 1×1 convolutional layer. This represents a 3×3 depth convolutional layer. express Values, low-frequency information After a 1×1 convolutional layer, we obtain , This represents a 1×1 convolutional layer. express The key, low-frequency information After passing through a 1×1 convolutional layer and a 3×3 depthwise convolutional layer, we obtain... , This represents a 1×1 convolutional layer. This represents a 3×3 depth convolutional layer. express The value, and then for and Softmax normalization is performed to obtain , This represents the softmax activation function, and finally... and Element-wise multiplication yields high-frequency information after interaction. , Indicates element-wise multiplication, and Element-wise multiplication yields low-frequency information after interaction. , This indicates element-wise multiplication; S6. Fishery resource statistical image processing: Input the fishery resource statistical image to be processed into the color prior feature enhancement network to obtain the enhanced fishery resource statistical image.

2. The image enhancement processing method for fishery resource statistics according to claim 1, characterized in that, In S2, for the color prior feature enhancement network, the input is a fishery resource statistical image. H, W and C represent respectively The height, width, and channels are determined using discrete wavelet transform. Decomposed into high-frequency information and low-frequency information, , Represents discrete wavelet transform. This represents the low-frequency information of the input fishery resource statistics image, including the image's content and color information. This inputs high-frequency information from a fisheries resource statistical image, including detailed information on global structure and texture. Then, the high-frequency and low-frequency information are input into a feature enhancement module, which consists of a low-frequency information enhancement module, a high-frequency information enhancement module, and a frequency information interaction module. The low-frequency information... Obtained through the low-frequency information enhancement module High-frequency information After being obtained through the high-frequency information enhancement module , and The input to the frequency information interaction module enables information exchange between high-frequency and low-frequency information. , , This indicates the frequency information interaction module. This indicates low-frequency information following the interaction. Indicate the high-frequency information after the interaction, and then and Adding elements together yields , This indicates the addition of elements. and Adding elements together yields , This indicates the addition of elements. and As input to the next feature information enhancement module, it is obtained after passing through N feature information enhancement modules. and , and The final image is generated after inverse discrete wavelet transform. , This represents the inverse discrete wavelet transform.

Images