A water body image enhancement method for fishery environment dynamic monitoring

Abstract

The application provides a water body image enhancement method for dynamic monitoring of a fishery environment, and relates to the field of image enhancement.The application constructs a water body image enhancement model, adopts a self-adaptive feature enhancement module, enhances high-frequency information of a key area of the water body image through frequency domain feature learning and information extraction, and combines a nonlinear enhancement strategy to improve image quality; a detail recovery module is introduced, a local feature similarity of pixels is taken as a basis to dynamically adjust an enhancement factor, so that the image edge is clearer; a global information understanding module calculates global dependence weights between pixels, fuses long-distance pixel information, so that the overall information of the water body image is more balanced, and information loss caused by local enhancement is inhibited; and the application is applied to the field of fishery environment monitoring and provides a clear image enhancement scheme.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of image enhancement, and particularly relates to a water body image enhancement method for dynamic monitoring of fishery environment. BACKGROUND

[0002] In fishery environment monitoring, the dynamic changes of water body directly affect the growth and ecological balance of fish, and timely and accurate acquisition of water body image information is crucial for fishery production management, ecological protection and water quality monitoring. Traditional fishery environment monitoring mainly relies on manual patrol and water quality sensors, but manual monitoring has problems such as poor real-time performance, limited coverage and susceptibility to subjective errors, while water quality sensors can provide physical and chemical parameters but cannot directly analyze fish activity, pollutant distribution and environmental changes in water body. Therefore, water body monitoring methods based on underwater cameras and computer vision have gradually become a research hotspot. However, due to the complexity of water environment, including light changes, light scattering and attenuation, low contrast caused by turbid water quality, scale changes of target objects and motion blur, traditional image processing methods are difficult to be directly applied to water body monitoring scenes, and existing water body image enhancement techniques still have limitations in adapting to different water environments, preserving target details and reducing noise interference.

[0003] Current water body image processing methods include histogram equalization, Retinex theory, and deep learning-based image enhancement techniques. Histogram equalization can improve the contrast of images, but it is easy to cause local overexposure or loss of details. Retinex method enhances the brightness of low light area by simulating the human eye visual model, which is suitable for some water environments, but it is easy to produce artifacts in high noise conditions. Deep learning methods, such as convolutional neural network (CNN), generative adversarial network (GAN) or visual Transformer-based enhancement methods, can obtain strong feature extraction ability through large amount of data training, but due to the difficulty of underwater image data labeling in fishery environment, deep learning methods have strong dependence on high-quality training data and high computational cost, which is difficult to be deployed in real time in fishery monitoring scenarios. In addition, existing methods are often limited to local pixel enhancement and cannot effectively combine global information, resulting in unsatisfactory segmentation effect of water body targets and background, affecting the monitoring accuracy. SUMMARY

[0004] The present application provides a water body image enhancement method for dynamic monitoring of fishery environment, aiming to improve the quality of water body images and enhance the details and clarity of images through deep learning models and feature extraction enhancement techniques.

[0005] The present application aims to provide a water body image enhancement method for dynamic monitoring of fishery environment, which specifically includes the following steps.

[0006] S1, collect the fishery environment of dynamic monitoring shooting, pretreatment and image enhancement, and make water body image dataset.

[0007] S2, construct an adaptive image feature enhancement module through local geometric feature learning and adaptive information extraction, introduce an enhancement factor of dynamic adjustment frequency domain to strengthen the high frequency information of the key area of the image, and process the water body image through nonlinear information enhancement, and finally perform weighted fusion.

[0008] S3, construct a detail recovery module according to the local feature similarity of each pixel point of the water body image, introduce a dynamic adjustment factor, and use the adjustment factor to weight the local feature to enhance the detail information.

[0009] S4, introduce a global dependence weight to construct a global information understanding module, fuse the information of different pixels through the global dependence weight, and realize adaptive global information aggregation.

[0010] S5, construct a water body image enhancement model, the water body image enhancement model comprises input, image embedding, adaptive image feature enhancement module, detail recovery module, global information understanding module, full connection layer and output.

[0011] S6, water body image enhancement model training and testing, using the water body image dataset to train the model.

[0012] Preferably, in the step S1, the water body pictures of the fishery environment under the dynamic environment are shot by using the unmanned aerial vehicle, the multi-view cameras on the water surface and underwater are configured, the underwater and water surface images are collected, the images cover various fishery environments including fresh water, sea water, lake and river, the water body image is pretreated and image enhanced, the water body image dataset is made, and the dataset is divided into training set, verification set and test set.

[0013] Preferably, in the step S2, the adaptive image feature enhancement module is constructed, which comprises the following steps:

[0014] Step S21, input fishery water body image features , , and respectively represent the height, width and channel number of the water body image feature , first calculate the local gradient of the image feature through the local geometric feature of the image, capture the texture and edge information in the image, and calculate the gradient of the water body image feature X in the horizontal direction and the vertical direction:

[0015] ;

[0016] ;

[0017] wherein, represents the eigenvalue at coordinate (x, y), and respectively represent the local gradients of the water body image feature in x and y directions, representing the edge changes of the image;

[0018] According to the local gradient, the local geometric information of the image feature is calculated :

[0019] ;

[0020] Step S22, in order to apply different convolution kernels with different intensities in different image feature regions, the adaptive function is used to generate the convolution kernel according to the extracted local geometric information,

[0021] Adaptive convolution kernel is dynamically generated through the local geometric information , and the generation formula is:

[0022] ;

[0023] wherein, is a globally learned weight parameter,

[0024] Then, the water body image feature is convolved by the generated adaptive convolution :

[0025] ;

[0026] wherein, is the water body image feature after the adaptive convolution operation, n is the size of the convolution kernel, k represents the horizontal direction offset, and l represents the vertical direction offset;

[0027] Step S23, using fast Fourier transform FFT to convert the water body image feature from the spatial domain to the frequency domain:

[0028] ;

[0029] is the representation of the image feature in the frequency domain, containing all the frequency information of the image,

[0030] Define a dynamic adjustment enhancement factor in the frequency domain to control the enhancement intensity of the high frequency part in the frequency domain, and the calculation formula is:

[0031] ;

[0032] wherein, is a weight learned through training, controlling the influence degree of the gradient, is the local gradient of the image feature, representing the intensity of change of the image at this position, b is the offset term, is the activation function,

[0033] using an enhancement factor to dynamically adjust the high-frequency information enhancement strength in the frequency domain, and the specific enhancement process is:

[0034] ;

[0035] Then use the inverse Fourier transform IFFT to restore the enhanced result in the frequency domain to the spatial domain:

[0036] ;

[0037] Step S24, in order to further enhance the contrast and details of the water body image features, the RELU nonlinear enhancement method is adopted, and the details in the image are enhanced through the activation function:

[0038] ;

[0039] Finally, the frequency domain enhanced image features, the convolution image features and the nonlinear enhanced image features are fused to obtain the final enhanced image features, and weighted fusion is used to balance the contribution of different enhanced features, and the specific formula is:

[0040] ;

[0041] and is the weight coefficient learned in the training process, , and control the contribution of the spatial domain convolution feature, the frequency domain enhanced feature and the nonlinear enhanced feature respectively.

[0042] Preferably, in step S2, for the adaptive image feature enhancement module, through local geometric feature learning and adaptive information extraction, the enhancement strategy of the image is dynamically adjusted, the high-frequency information of the key area is strengthened, and the detail preservation ability of the image is improved; in the frequency domain, the enhancement factor of the dynamic adjustment of the frequency domain is introduced, the high-frequency information is enhanced, and the edges, details and structures of the image are more clear; the nonlinear information enhancement method is used to further optimize the image contrast and enhance the distinguishability of complex texture structure; through weighted fusion of the features from the spatial domain, the frequency domain and the nonlinear enhanced features, the water body image can maintain the natural structure while obtaining better clarity and visual quality, and the robustness and adaptability to environmental changes are improved.

[0043] Preferably, in step S3, a detail recovery module is constructed, specifically including the following steps:

[0044] adopting local neighborhood calculating The similarity of the pixel at point and the neighborhood pixels is calculated as follows:

[0045] ;

[0046] In the formula, is the value of the input image feature at pixel position , represents a 3x3 local neighborhood centered on the pixel point , represents the feature value at (i,j) in the field, is an exponential decay coefficient that controls the similarity, which determines the weight of the pixels away from the center pixel, is used to measure the similarity, and the larger the value, the more similar the pixel and the neighborhood pixel are;

[0047] According to the local feature similarity, a dynamic adjustment factor is defined to adaptively adjust the features of different regions, so that the details are clearer, while avoiding the noise caused by over-enhancement. The adjustment factor is calculated as follows:

[0048] ;

[0049] In the formula, represents the dynamic adjustment factor of the pixel , the adjustment factor value is between , represents a smoothing factor that controls the effective range of the dynamic adjustment factor;

[0050] The adjustment factor is used to adjust the local features by weighting to enhance important details in the water image:

[0051] ;

[0052] In the formula, represents the pixel value of the adjusted local feature, is the weighted mean value of the local neighborhood, The calculation formula is as follows:

[0053] .

[0054] Preferably, in step S3, for the detail recovery module, based on the local feature similarity of each pixel point of the water body image, the enhancement strategy of the image is adaptively adjusted to improve the retention ability of the detail information; Specifically, by calculating the similarity of the pixel point and its neighborhood, a local feature similarity measure (LFS) is constructed to measure the correlation of the pixel and the neighborhood, and on this basis, a dynamic adjustment factor (DAF) is introduced to adaptively adjust the feature enhancement intensity of different regions; By using the dynamic adjustment factor to weight and fuse the local features, the key details can be strengthened while maintaining the overall consistency of the image, thereby improving the clarity and structural information fidelity of the image, and reducing the noise interference caused by over-enhancement.

[0055] Preferably, in step S4, a global information understanding module is constructed, specifically including the following steps:

[0056] For any two pixels and , the feature similarity is calculated, and the calculation method is:

[0057] ;

[0058] In the formula, represents the Euclidean feature distance between the two pixels, which measures their similarity, controls the decay degree of similarity,

[0059] Then, the global dependence weight is calculated according to the similarity score:

[0060] ;

[0061] The global dependence weight is used to adjust the feature of each pixel to obtain the enhanced water body image feature , so that more information can be perceived in the global range:

[0062] .

[0063] Preferably, in step S4, for the global information understanding module, the feature similarity between any pixel pair is first calculated, and then the global dependence weight is calculated based on the similarity distribution, which is used to weight and fuse the pixel information; In this way, the global information can be effectively aggregated, so that each pixel point not only depends on its local feature, but also fully utilizes the global information to improve the overall consistency of the image.

[0064] Compared with the prior art, the present application has the following technical effects:

[0065] Through local geometric feature learning and adaptive information extraction, the high-frequency information of the water body image can be enhanced, and the contrast and image definition can be improved; a dynamic adjustment frequency domain enhancement factor is introduced, and the key regions of the image are adaptively enhanced, so that the structural information of the image is effectively improved, and noise amplification caused by over-enhancement is avoided; in addition, the application combines a global information understanding module, calculates a global dependence weight, realizes adaptive fusion of different pixel information, ensures global consistency of the image, and improves overall perception quality. BRIEF DESCRIPTION OF DRAWINGS

[0066] Figure 1 is a water body image enhancement method flowchart provided by the application for dynamic monitoring of fishery environment.

[0067] Figure 2 is a structure diagram of an adaptive image feature enhancement module provided by the application.

[0068] Figure 3 is a structure diagram of a detail recovery module provided by the application.

[0069] Figure 4 is a structure diagram of a global information understanding module provided by the application.

[0070] Figure 5 is an effect diagram before water body image enhancement provided by the application.

[0071] Figure 6 is an effect diagram after water body image enhancement provided by the application. DETAILED DESCRIPTION

[0072] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0073] Please refer to the accompanying Figures 1-6 , the application provides a water body image enhancement method for dynamic monitoring of fishery environment.

[0074] As shown in the flowchart of the accompanying Figure 1 , the application provides a water body image enhancement method for dynamic monitoring of fishery environment, and the specific implementation manner includes the following steps:

[0075] S1, collect the fishery environment shot by dynamic monitoring, perform pretreatment and image enhancement, and make water body image dataset.

[0076] Further, in the S1 step, the unmanned aerial vehicle is used to shoot the fishery environment water body pictures in the dynamic environment, the multi-view camera on the water surface and underwater is configured, the underwater and water surface images are collected, the images cover various fishery environments including fresh water, sea water, lakes and rivers, the water body images are preprocessed and image enhancement operations are performed, the water body image dataset is made, and the dataset is divided into a training set, a validation set and a test set.

[0077] S2, an adaptive image feature enhancement module is constructed by innovative local geometric feature learning and adaptive information extraction, an enhancement factor of a dynamic adjustment frequency domain is introduced to strengthen the high-frequency information of the key region of the image, and the water body image is processed through nonlinear information enhancement, and finally weighted fusion is performed.

[0078] Further, in the S2 step, the adaptive image feature enhancement module is constructed, as shown in Figure 2 , specifically including the following steps:

[0079] Step S21, input the fishery water body image features , , and respectively represent the height, width and channel number of the water body image features , first, through the local geometric features of the image, the local gradient of the image features is calculated, the texture and edge information in the image is captured, and the gradient of the water body image features X in the horizontal direction and the vertical direction is calculated:

[0080] ;

[0081] ;

[0082] wherein represents the feature value at the coordinates (x, y), and respectively represent the local gradient of the water body image features in the x and y directions, and represent the edge change of the image;

[0083] According to the local gradient, the local geometric information of the image features is calculated:

[0084] ;

[0085] Step S22, in order to apply different intensity convolution kernels in different image feature regions, according to the extracted local geometric information, the convolution kernel is generated through an adaptive function,

[0086] the adaptive convolution kernel is dynamically generated through the local geometric information , and the generation formula is:

[0087] ;

[0088] wherein, is a global learning weight parameter, in the embodiment, the initial value is set to 0.5, the value range is 0 to 1, and the Adam optimization algorithm is used for updating.

[0089] Then the generated adaptive convolution is used to convolve the water body image features:

[0090] ;

[0091] wherein, is the water body image feature after adaptive convolution operation, n is the size of the convolution kernel, k represents the horizontal offset, and l represents the vertical offset;

[0092] Step S23, using fast Fourier transform (FFT) to convert the water body image feature from the spatial domain to the frequency domain:

[0093] ;

[0094] is the representation of the image feature in the frequency domain, containing all the frequency information of the image,

[0095] The dynamic adjustment of the enhancement factor in the frequency domain is defined to control the enhancement intensity of the high frequency part in the frequency domain, and the calculation formula is:

[0096] ;

[0097] wherein, is an activation function, is a weight learned through training, controlling the influence degree of the gradient, in the embodiment, the initial value is set to 1, the value range is -1 to 1, the normal distribution is used for initialization, and the Adam optimization algorithm is used for updating, is the local gradient of the image feature, indicating the change intensity of the image at this position, b is the offset term, the initial value of b is set to 0, the value range is -0.2 to 0.2, the uniform distribution is used for initialization, and the Adam optimization algorithm is used for updating,

[0098] The enhancement factor is used to dynamically adjust the high frequency information enhancement intensity in the frequency domain, and the specific enhancement process is:

[0099] ;

[0100] Then, the frequency domain enhancement result is restored to the spatial domain using the inverse Fourier transform (IFFT):

[0101] ;

[0102] Step S24: To further enhance the contrast and detail of the water body image features, the ReLU nonlinear enhancement method is used. This activation function enhances the details in the image.

[0103] ;

[0104] Finally, the frequency-domain enhanced image features are fused with the convolutional image features and the nonlinearly enhanced image features to obtain the final enhanced image features. Weighted fusion is used to balance the contributions of different enhanced features, and the specific formula is as follows:

[0105] ;

[0106] and These are the weight coefficients learned during the training process. , and In this embodiment, the contributions of spatial domain convolutional features, frequency domain enhancement features, and nonlinear enhancement features are controlled separately. The initial value is set to 0.4. The initial value is set to 0.4. During the update process, and The sum of the two numbers is less than or equal to 1.

[0107] S3. Construct a detail restoration module based on the local feature similarity of each pixel in the water image, introduce a dynamic adjustment factor, and use the adjustment factor to weight the local features to enhance the detail information.

[0108] Furthermore, in step S3, a detailed recovery module is constructed, such as... Figure 3 As shown, the specific steps include:

[0109] Using local neighborhood calculate The similarity between a pixel at a given point and its neighboring pixels is calculated using the following formula:

[0110] ;

[0111] In the formula, For input image features at pixel locations The value at that location, Represented by pixels A 3×3 local neighborhood centered on the center. express Eigenvalue at (i,j) in the field, is the exponential decay coefficient for controlling similarity, determines the weight of pixels far away from the center pixel, in this embodiment, is set to 4, is used to measure similarity, the greater the value, the more similar the pixel is to the neighboring pixels;

[0112] According to the local feature similarity, a dynamic adjustment factor is defined for adaptively adjusting the features of different regions, so that the details are clearer, while avoiding the noise caused by over-enhancement, and the adjustment factor calculation formula is:

[0113] ;

[0114] In the formula, represents the dynamic adjustment factor of the pixel , the adjustment factor value is between , represents a smoothing factor, controls the action range of the dynamic adjustment factor, in this embodiment, the value of is set to 0.5;

[0115] The adjustment factor is used to adjust the local features by weighting, to enhance important details in the water body image:

[0116] ;

[0117] In the formula, represents the pixel value of the adjusted local feature, is the weighted mean value of the local neighborhood, the calculation formula is:

[0118] .

[0119] S4, introduce global dependence weight to construct global information understanding module, through global dependence weight, the information of different pixels is fused, adaptive global information aggregation is realized.

[0120] Further, in the S4 step, the global information understanding module is constructed, as shown in Figure 4 , specifically including the following steps:

[0121] For any two pixels and , calculate their feature similarity , the calculation method is:

[0122] ;

[0123] In the formula, represents the Euclidean feature distance between two pixels, measuring their similarity, controlling the degree of attenuation of similarity, in this embodiment, the value of is set to 2.5,

[0124] Then calculate the global dependency weight according to the similarity score :

[0125] ;

[0126] Use the global dependency weight to adjust the features of each pixel to get the enhanced water image features , so that more information can be perceived in the global range:

[0127] .

[0128] S5, construct a water image enhancement model, which includes input, image embedding, adaptive image feature enhancement module, detail recovery module, global information understanding module, fully connected layer and output.

[0129] Further, for the water image enhancement model, input the low resolution water image , , , and respectively represent the height, width and channel number of , in this embodiment, H=640, W=640, C=3, that is, the input image is , input to the image embedding, which contains 3x3 convolution, to get the low resolution water image feature , input to the adaptive image feature enhancement module to get , input to the detail recovery module to get , input to the global information understanding module to get the enhanced water image feature , input to 3x3 convolution for processing, and the processed water image feature is input to the fully connected layer, and finally output to get , .

[0130] S6, water image enhancement model training and testing, using the prepared water image dataset to train the model.

[0131] Furthermore, in step S6, the water image enhancement model is coded using the PyCharm application and the Python language, and trained using the PyTorch framework. The model is trained on a low-resolution water image with a resolution of 640×640×3 as input, and the model is trained from scratch for 200 epochs.

[0132] Furthermore, such as Figure 5 and Figure 6 As shown, Figure 5 Low-resolution water images were displayed. Figure 6 The image showcases a high-resolution water image obtained through water image enhancement model processing. It can be seen that the image clarity and detail in the water surface ripples and fish schools have been significantly improved, with more distinct details.

[0133] 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. A method for enhancing water body images for dynamic monitoring of the fishery environment, characterized in that, Includes the following steps: S1. Collect dynamic monitoring images of the fishery environment, perform preprocessing and image enhancement, and create a water body image dataset; S2. An adaptive image feature enhancement module is constructed through local geometric feature learning and adaptive information extraction. That is, the local gradient of the image features is calculated through the local geometric features of the image, and then the local geometric information of the image features is calculated based on the local gradient. Then, a convolution kernel is generated through an adaptive function for convolution processing. Then, a dynamically adjusted enhancement factor in the frequency domain is introduced to enhance the high-frequency information of key areas of the image. The water image is enhanced through nonlinear information enhancement processing, and finally, weighted fusion is performed. S3. Construct a detail restoration module based on the local feature similarity of each pixel in the water image, introduce a dynamic adjustment factor, and use the adjustment factor to weight the local features to enhance the detail information. The specific method for designing the detail restoration module is as follows: Using local neighborhood calculate The similarity between a pixel at a given point and its neighboring pixels is calculated using the following formula: ； In the formula, For input image features at pixel locations The value at that location, Represented by pixels A 3×3 local neighborhood centered on the center. express The eigenvalue at point (i,j) in the neighborhood. To control the exponential decay coefficient of similarity; Based on local feature similarity, a dynamic adjustment factor is defined, and the formula for calculating the adjustment factor is as follows: ； In the formula, Represents pixels The dynamic adjustment factor, adjustment factor Value at between, This represents the smoothing factor, which controls the scope of the dynamic adjustment factor. Use adjustment factor Weighted adjustments are made to local features: ； In the formula, This represents the pixel value after adjusting for local features. It is the weighted mean of the local neighborhood. The calculation formula is: ； S4. Introduce a global dependency weight to construct a global information understanding module. Through the global dependency weight, information from different pixels is fused to achieve adaptive global information aggregation. S5. Construct a water body image enhancement model, which includes an input, an image embedding, an adaptive image feature enhancement module, a detail restoration module, a global information understanding module, a fully connected layer, and an output. S6. Training and testing of the water image enhancement model: The model is trained using the created water image dataset.

2. The water image enhancement method for dynamic monitoring of the fishery environment according to claim 1, characterized in that, In step S1, the method for creating the water body image dataset is as follows: Using drones to capture images of fishery waters in dynamic environments, multi-view cameras were configured for both surface and underwater use to collect images covering various fishery environments, including freshwater, seawater, lakes, and rivers. The water images were preprocessed and enhanced to create a water image dataset, which was then divided into training, validation, and test sets.

3. The water image enhancement method for dynamic monitoring of the fishery environment according to claim 2, characterized in that, In step S2, the adaptive image feature enhancement module is constructed using the following method: S21. Input fishery water body image features , , and These represent the features of water bodies in the image. The height, width, and number of channels are first determined by calculating the local gradient of image features using local geometric features, capturing texture and edge information in the image, and then calculating the gradient of water body image feature X in the horizontal and vertical directions. ； ； in, This represents the eigenvalue at coordinates (x, y). and These represent the local gradients of water body image features in the x and y directions, respectively, and represent the edge changes of the image; Based on the local gradient, calculate the local geometric information of image features. : ； S22. Based on the extracted local geometric information, generate convolution kernels using an adaptive function. Adaptive convolution kernel Through local geometric information It is dynamically generated, and the generation formula is: ； in, These are the weight parameters for global learning. Then, through the generated adaptive convolution... Convolution processing is performed on the features of the water body image: ； in, These are the water image features after adaptive convolution operation, where n is the size of the convolution kernel, k represents the horizontal offset, and l represents the vertical offset. S23. Use Fast Fourier Transform (FFT) to extract water body image features. Transformation from the spatial domain to the frequency domain: ； It is the frequency domain representation of image features, containing all frequency information of the image. Define the enhancement factor for dynamically adjusting the frequency domain The enhancement intensity in the high-frequency range of the frequency domain is controlled by the following formula: ； in, These are weights learned through training. Here, b represents the local gradient of the image features, and b is the offset term. For activation function, Use enhancement factor To dynamically adjust the enhancement intensity of high-frequency information in the frequency domain, the specific enhancement process is as follows: ； Then, the frequency domain enhancement result is restored to the spatial domain using the inverse Fourier transform (IFFT): ； S24. The ReLU nonlinear enhancement method is used to enhance the details in the image through this activation function: ； Finally, the frequency-domain enhanced image features are fused with the convolutional image features and the nonlinearly enhanced image features to obtain the final enhanced image features. Weighted fusion is used to balance the contributions of different enhanced features, and the specific formula is as follows: ； and These are the weight coefficients learned during the training process.

4. The water image enhancement method for dynamic monitoring of the fishery environment according to claim 3, characterized in that, In step S4, a global information understanding module is designed, and the specific method is as follows: For any two pixels and Calculate their feature similarity The calculation method is as follows: ； In the formula, Controlling the degree of similarity decay, Then, the global dependency weight is calculated based on the similarity score. : ； Use global dependency weights To adjust the features of each pixel, the enhanced water body image features are obtained. : 。

Images