Remote sensing image water body extraction method based on multi-scale feature fusion of compression pyramid gating unit

Abstract

The present application relates to a kind of remote sensing image water body extraction methods based on compression pyramid gate unit's multiscale feature fusion, belong to remote sensing technical field.It includes: obtaining water body remote sensing image dataset, data set is preprocessed to carry out data set division, and carries out data enhancement processing;Using CNN hybrid Transformer encoder to extract multiple scale feature maps from image;Different scale feature maps are added to multiscale skip connection module respectively, and multiple scale feature information is extracted;Multiple scale feature information is fused, and the fused feature is multiplied by the weight of each scale feature to obtain the final multiscale adaptive receptive field selection result feature map;Cross-entropy loss is used to train the above process, and after training, it is used for remote sensing image water body extraction.The present application can effectively improve the extraction ability of water body of different shape and size and the distinguishing ability of water body information and non-water body information.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing technology and relates to a method for extracting water bodies from remote sensing images based on multi-scale feature fusion of compressed pyramid gating units. Background Technology

[0002] With the continuous development of remote sensing technology and the widespread application of high-resolution remote sensing imagery, water body extraction from high-resolution remote sensing images has attracted widespread attention in the remote sensing community. As one of humanity's most important resources, automated water body extraction is of great significance for understanding the current state of water resources, assisting countries in rationally planning and managing water resources, helping to predict floods or droughts and other disasters, and analyzing human activities. Moreover, water body recognition in remote sensing images has always been an important part of scientific progress. In recent years, semantic segmentation methods based on deep learning have been widely applied in water body extraction. However, many shortcomings still need to be addressed. For example, there are problems such as blurred water body boundary information, susceptibility to background occlusion such as buildings and mountain shadows, and the tendency for false positives and false negatives in thin, elongated water flows. Many deep learning-based methods have emerged for water body extraction, addressing issues such as irregular sizes and shapes, small water bodies, and occlusion by non-water body shadows. For example, PSPNet uses a pyramid pooling structure to effectively integrate multi-scale contextual information, but it has some limitations when processing small targets or detailed features.

[0003] In recent years, with the rapid development of convolutional neural networks (CNNs), significant breakthroughs have been achieved in water extraction tasks. The introduction of fully convolutional networks (FCNs) was the first to apply a fully convolutional network structure to the field of water extraction, which has attracted the attention of many researchers who are dedicated to designing encoder-decoder architectures to achieve higher accuracy in water extraction. CNN-based water extraction frameworks have emerged continuously, and these methods have improved extraction accuracy to some extent. However, these methods have certain limitations in water convolution operations, making it difficult to effectively model global contextual information. Due to the huge differences in the shape and size of water bodies, especially some extremely irregular and small water bodies, and the complexity of the surrounding environment (such as occlusion by mountains or building shadows), accurate inference of local contextual information becomes increasingly difficult. Therefore, how to overcome these challenges remains an important problem to be solved in the field of water extraction. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a method for extracting water bodies from remote sensing images based on multi-scale feature fusion of compressed pyramid gating units.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for extracting water bodies from remote sensing images based on multi-scale feature fusion using compressed pyramid gating units, comprising the following steps:

[0007] S1. Obtain the water body remote sensing image dataset, preprocess the dataset, divide it into datasets, and perform data augmentation processing.

[0008] S2. Use a CNN hybrid Transformer encoder to extract feature maps of multiple scales from the image;

[0009] S3. Add feature maps of different scales to the multi-scale skip connection module to extract feature information at multiple scales;

[0010] S4. Fuse the feature information from multiple scales, and multiply the fused feature with the features of each scale by weights to obtain the final multi-scale adaptive receptive field selection result feature map.

[0011] S5. The above process is trained using cross-entropy loss, and the trained data is then used for water body extraction from remote sensing images.

[0012] Furthermore, in step S1, the preprocessing of the dataset includes cropping, labeling, and binarizing the label pixels of the labeled images;

[0013] The dataset partitioning includes dividing the dataset into training, validation, and test sets according to a preset ratio;

[0014] Data augmentation of the dataset includes: random scaling, random horizontal and vertical flipping, random Gaussian noise, contrast-limited adaptive histogram equalization, random rotation, and random cropping.

[0015] Furthermore, in step S2, at least three CNN and Transformer networks are set in the CNN hybrid Transformer encoder to extract feature maps of three scales from three stages. After passing through each stage in sequence, three feature maps with resolutions of {1 / 4, 1 / 8, 1 / 16} of the original image are obtained, wherein the number of feature map channels C1 to C3 in each stage are {64, 128, 256} respectively.

[0016] Each stage includes a CNN layer and a Transformer layer; each stage extracts global features by first feeding the input features into the Transformer layer, and then feeding them into the CNN layer to extract local features, and so on.

[0017] Each Transformer layer includes a W-MSA Block and an SW-MSA Block; the W-MSA Block includes two layer normalizations, a multilayer perceptron, and a W-MSA, and the SW-MSA Block includes two layer normalizations, a multilayer perceptron, and an SW-MSA.

[0018] Each CNN layer includes two 3x3 convolutions, which take the results extracted by the Transformer and feed them into the convolutions to further extract local features.

[0019] Furthermore, in step S3, the multi-scale skip connection module includes several convolutional kernels of different sizes. The feature map of each stage is input into convolutions with kernels of different sizes to obtain feature maps of different scales. The feature maps of different scales are concatenated and stitched together according to the channel dimension to obtain the output of the multi-scale skip connection module, which includes:

[0020] S31. Perform several parallel convolution operations on the feature maps C1, C2, and C3 at different levels:

[0021] x1=Conv kxk (C1), k = 1, 3, 5, 7

[0022] x2=Conv kxk (C2), k = 1, 3, 5, 7

[0023] x3=Conv kxk (C3), k = 1, 3, 5, 7

[0024] Conv kxk () represents a convolution with a kernel size of k;

[0025] S32. Concatenate and fuse the convolution results x1, x2, and x3 along the channel dimension to obtain a feature map x containing multi-scale information. k :

[0026] x k =Cat[x1, x2, x3]

[0027] Cat[] indicates cascading splicing according to channel dimension.

[0028] Furthermore, in step S4, for the obtained feature maps x1, x2, x3, a 1x1 convolution kernel is used to process them to obtain y1, y2, y3 of the same dimension. Then, y1, y2, y3 are cascaded and fused according to the channel dimension to obtain a feature map yk containing multi-scale information.

[0029] y1=Conv 1x1 (x1)

[0030] y2=Conv 1x1 (x2)

[0031] y3=Conv 1x1 (x3)

[0032] y k=Cat[y1, y2, y3]

[0033] The feature map yk is input into the compressed pyramid gated unit module. After three sequential large kernel selection convolutions, three features z1, z2, and z3 of the three adaptive receptive fields are obtained, and these features are concatenated to obtain the feature map F. c :

[0034] z1 = Conv 3×3 (y k )

[0035] z2=Conv 5×5 (z1)

[0036] z3 = Conv 7×7 (z2)

[0037] F c =Cat[z1, z2, z3]

[0038] Then, F c It is reconstructed into matrix F through 1x1 convolution transpose and reshaping operations. H ∈R H×1 sum matrix F W ∈R 1×W :

[0039] F H =Conv 1×1 (reshape(Conv 1×1 (F C )))

[0040] F W =Conv 1×1 (reshape(Conv 1×1 (F C )))

[0041] In F H With F W Perform matrix multiplication and then reshape the matrix to obtain the eigenvector F. L ∈R 1×L L = H × W:

[0042] F L =reshape(F H ⊙F W )

[0043] Then use 1×1 convolution and reshaping operations to transform the feature map F c Mapped to matrix F A ∈R C×L :

[0044] F A=reshape(Conv 1×1 (F C ))

[0045] F L and F A Perform matrix multiplication between them, then reshape the vector to obtain the channel attention weight vector A. map ∈R C ×1×1 :

[0046] A map =reshape(F L ⊙F A )

[0047] Perform 1×1 convolutions between inputs, layer normalization, and apply the sigmoid activation function to obtain the weights W:

[0048] W=δ(Conv 1×1 (LN(A map )))

[0049] W and fusion features x k Element-wise multiplication yields the final multi-scale adaptive receptive field selection result feature map F. out :

[0050]

[0051] In the above process, reshape represents the reshaping operation, δ represents the sigmoid activation function, LN represents layer normalization, and ⊙ represents matrix multiplication. This indicates element-wise multiplication.

[0052] Furthermore, in step S5, the AdamW optimizer and poly learning rate adjustment strategy are used to calculate the loss between the predicted and ground truth maps using the cross-entropy loss function. The cross-entropy loss calculation formula is as follows:

[0053]

[0054] Where N represents the total number of pixels in the image, y i ∈{0,1} represents the true label of pixel i. This represents the probability that pixel i belongs to category 1.

[0055] The beneficial effects of this invention are as follows:

[0056] This invention employs an encoder combining CNN and Transformer methods to extract global and local contextual information. A multi-scale skip connection module effectively captures features at different scales, improving segmentation results through contextual information. A multi-level feature fusion module and a compressed pyramid gating unit effectively enhance the extraction capability for water bodies of different shapes and sizes, as well as the ability to distinguish between water and non-water information. The compressed pyramid gating unit significantly improves the model's feature representation and generalization capabilities through multi-scale information fusion, adaptive receptive field selection, channel attention mechanisms, and weight adjustment and feature optimization. This design helps the model better capture key features in complex scenes, improving the accuracy of recognition or classification.

[0057] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0058] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0059] Figure 1 This is a schematic diagram of the overall architecture of the remote sensing image water body extraction method of the present invention;

[0060] Figure 2 This is a schematic diagram of the structure of the multi-scale jump connection module of the present invention;

[0061] Figure 3 This is a schematic diagram of the structure of the compression pyramid gating unit of the present invention. Detailed Implementation

[0062] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0063] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0064] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0065] Please see Figures 1-3 This paper proposes a method for water body extraction from remote sensing images based on multi-scale feature fusion of compressed pyramid gating units, in order to solve the problems of irregularly shaped and sized water bodies and occlusion of non-water areas in current water body extraction methods.

[0066] Example

[0067] This embodiment presents a method for water body extraction from remote sensing images based on multi-scale feature fusion using compressed pyramid gating units, such as... Figure 1 As shown, the conclusion includes the following steps:

[0068] S1. Obtain the water body remote sensing image dataset, preprocess the dataset, divide it into datasets, and perform data augmentation processing.

[0069] S2. Use a CNN hybrid Transformer encoder to extract feature maps of multiple scales from the image;

[0070] S3. Add feature maps of different scales to the multi-scale skip connection module to extract feature information at multiple scales;

[0071] S4. Fuse the feature information from multiple scales, and multiply the fused feature with the features of each scale by weights to obtain the final multi-scale adaptive receptive field selection result feature map.

[0072] S5. The above process is trained using cross-entropy loss, and the trained data is then used for water body extraction from remote sensing images.

[0073] In step S1 of this embodiment, the preprocessing of the dataset includes cropping, labeling, and converting the label pixels of the labeled images to 0 or 1. This embodiment uses an overlapping method for cropping, using a sliding window with a stride of 128 to crop the image to a size of 256×256.

[0074] The preprocessed dataset was used to construct training, validation, and test sets in an 8:1:1 ratio. During training, data augmentation was applied to the images, including random scaling, random horizontal and vertical flipping, random Gaussian noise, contrast-limited adaptive histogram equalization, random rotation, and random cropping. The random scaling ratio was 0.5-2.0. Random horizontal and vertical flipping, random Gaussian noise, contrast-limited adaptive histogram equalization, and random rotation were performed with a 50% probability. The random rotation angle was within the range of (-180, 180). Finally, the processed images were randomly cropped to a size of 256×256.

[0075] Specifically, the original 5000×5000 images of the Kaggle water dataset are cropped to 256×256 using a sliding window with a stride of 128. The dataset is divided into three subsets: training set, validation set, and test set. The pixel values ​​of the water regions in the labeled images of the dataset are set to 1, and the values ​​of other pixels are set to 0.

[0076] In step S2 of this embodiment, a convolutional neural network (CNN) and a Transformer module are combined to perform preliminary feature map extraction. Although CNN has excellent local feature extraction capabilities and has brought a qualitative breakthrough to water body extraction tasks, its ability to control global modeling is still lacking. With the advent of the Transformer, this problem has been improved. The Transformer, with its unique self-attention mechanism, can effectively capture global contextual information, overcoming the limitations of CNN in handling long-distance dependencies. This is particularly important for water body extraction tasks, because the shape and boundary of water bodies may be complexly intertwined with the surrounding environment, and local features are insufficient for accurate identification. Even more exciting is that the mechanism of mixing CNN with Transformer enables the model to capture details and understand global relationships, thereby improving task performance. Moreover, the model can more effectively integrate information at different scales, which is crucial for water body extraction tasks, enhancing the recognition of water body boundary information and improving the model's accuracy and robustness.

[0077] Specifically, the CNN-Transformer hybrid encoder in this embodiment sets up at least three CNN and Transformer networks to extract feature maps at three scales from three stages, with each stage including a CNN layer and a Transformer layer;

[0078] Each stage involves first feeding the input features into a Transformer layer to extract global features, and then feeding them into a CNN layer to extract local features. This process is repeated step-by-step.

[0079] Each Transformer Block includes a W-MSA Block and an SW-MSA Block; the W-MSA Block includes two layer normalizations, a multilayer perceptron, and a W-MSA, and the SW-MSA Block includes two layer normalizations, a multilayer perceptron, and an SW-MSA.

[0080] Each CNN layer includes two 3x3 convolutions. The results extracted by the Transformer are fed into these two convolutions to further extract local features.

[0081] In the above process, the image patch is input into the CNN hybrid Transformer encoder. After passing through each stage, three feature maps with resolutions of {1 / 4, 1 / 8, 1 / 16} of the original image are obtained. The number of channels C1 to C3 of the feature map in each stage are {64, 128, 256}, respectively.

[0082] In step S3 of this embodiment, a multi-scale skip connection module is used to extract features at different scales. In the field of water body extraction, the extraction of small water bodies and the interference of background noise have always been troublesome problems. Accurately identifying small water bodies and effectively reducing background noise interference can greatly improve the accuracy of water body segmentation and the robustness of the model. Applying multi-scale feature extraction allows the model to focus on features of different scales and sizes simultaneously, indicating that small water bodies can be effectively identified at a smaller scale, thereby improving the model's recognition rate. The multi-scale mechanism can also acquire background information at different levels, thereby enhancing the model's ability to distinguish between water and non-water areas.

[0083] Specifically, such as Figure 2As shown, the multi-scale skip connection module works as follows: The feature maps from each stage are input into convolutions with kernel sizes of 1x1, 3x3, 5x5, and 7x7, respectively, resulting in four feature maps at different scales. These four feature maps are then concatenated along the channel dimension to obtain the final output of the module. To capture multi-scale information, the three different-level feature maps C1, C2, and C3 obtained from the above operations are first subjected to four parallel convolution operations, with kernel sizes of 1, 3, 5, and 7, respectively. Then, the results x1, x2, and x3 are concatenated and fused along the channel dimension to obtain a feature map x with rich multi-scale information. k The entire multi-scale jump connection module can be expressed by the formula:

[0084] x1=Conv kxk (C1), k = 1, 3, 5, 7

[0085] x2=Conv kxk (C2), k = 1, 3, 5, 7

[0086] x3=Conv kxk (C3), k = 1, 3, 5, 7

[0087] x k =Cat[x1, x2, x3]

[0088] Among them, Conv kxk () indicates a convolution with kernel k, and Cat[] indicates concatenation by channel dimension.

[0089] In step S4 of this embodiment, the feature information of the three different scales obtained in step S3 is fused and multiplied with the weights generated by the compression pyramid gating unit of the three scale features.

[0090] Dynamically adjusting weights based on the relevance of input features is a unique characteristic of the attention mechanism. This allows the model to automatically enhance features related to water bodies when focusing on small water bodies or excluding non-water body features, thereby improving the model's ability to extract small water bodies and suppress background noise.

[0091] For x1, x2, and x3 obtained in step S3, considering the different sizes, a 1x1 convolution kernel was selected to process them to obtain y1, y2, and y3 of the same dimension. Then, y1, y2, and y3 were cascaded and fused according to the channel dimension to obtain a feature map yk with rich multi-scale information.

[0092] y1=Conv 1x1 (x1)

[0093] y2=Conv 1x1 (x2)

[0094] y3=Conv 1x1 (x3)

[0095] y k =Cat[y1, y2, y3]

[0096] Then in the compressed pyramid gating unit module, such as Figure 3 As shown, feature map yk is processed by three consecutive large kernel selection convolutions to obtain three features z1, z2, and z3 of three adaptive receptive fields, and these features are then concatenated to obtain feature map F. c .

[0097] z1 = Conv 3×3 (y k )

[0098] z2=Conv 5×5 (z1)

[0099] z3 = Conv 7×7 (z2)

[0100] F c =Cat[z1, z2, z3]

[0101] After that, F c First, it is reconstructed into matrix F through 1x1 convolution transpose and reshaping operations. H ∈R H×1 Similarly, matrix F is obtained. W ∈R 1×W Then, a similar operation is used, employing 1×1 convolution and reshaping operations to transform the feature map F. c Mapped to matrix F A ∈R C×L (L=H×W), then, in F H With F W Perform matrix multiplication and then reshape the matrix to obtain the eigenvector F. L ∈R 1 ×L (L = H × W). Finally, F L and F A Perform matrix multiplication between them, then reshape the vector to obtain the channel attention weight vector A. map ∈R C×1×1 Then, 1×1 convolutions are performed between the inputs, layer normalization is applied, and a sigmoid activation function is applied to obtain the weights W. W is then combined with the fused feature x obtained in step S3. k Element-wise multiplication yields the final multi-scale adaptive receptive field selection result feature map F. out .

[0102] FH =Conv 1×1 (reshape(Conv 1×1 (F C )))

[0103] F W =Conv 1×1 (reshape(Conv 1×1 (F C )))

[0104] F L =reshape(F H ⊙F W )

[0105] F A =reshape(Conv 1×1 (F C ))

[0106] A map =reshape(F L ⊙F A )

[0107] W=δ(Conv 1×1 (LN(A map )))

[0108]

[0109] Where reshape represents the reshaping operation, δ represents the sigmoid activation function, LN represents layer normalization, and ⊙ represents matrix multiplication. This indicates element-wise multiplication.

[0110] In step S5 of this embodiment, the AdamW optimizer and poly learning rate adjustment strategy are used, and the cross-entropy loss function is used to calculate the loss between the predicted map and the real map, and the loss is reduced through backpropagation.

[0111] Specifically, batch training is employed with a batch size of 16. This invention uses the AdamW optimizer and a Poly learning rate adjustment strategy, with the learning rate set to 0.0006 and the weight decay set to 0.00025.

[0112] The loss calculation method is as follows: calculate the cross-entropy loss between the final predicted image obtained in S4 and the ground image. The cross-entropy loss calculation formula is as follows:

[0113]

[0114] Where N represents the total number of pixels in the image, y i ∈{0,1} represents the true label of pixel i. This represents the probability that pixel i belongs to category 1.

[0115] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for extracting water bodies from remote sensing images based on multi-scale feature fusion using compressed pyramid gating units, characterized in that: It includes the following steps: S1. Obtain the water body remote sensing image dataset, preprocess the dataset, divide it into datasets, and perform data augmentation processing. S2. Use a CNN hybrid Transformer encoder to extract feature maps of multiple scales from the image; In step S2, at least three CNN and Transformer networks are set in the CNN hybrid Transformer encoder to extract feature maps of three scales from three stages. After passing through each stage in sequence, three feature maps with resolutions of {1 / 4, 1 / 8, 1 / 16} of the original image are obtained, where the number of channels of the feature map in each stage are {64, 128, 256} respectively. Each stage includes a CNN layer and a Transformer layer; each stage extracts global features by first feeding the input features into the Transformer layer, and then feeding them into the CNN layer to extract local features, and so on. Each Transformer layer includes a W-MSA Block and an SW-MSA Block; the W-MSA Block includes two layer normalizations, a multilayer perceptron, and a W-MSA, and the SW-MSA Block includes two layer normalizations, a multilayer perceptron, and an SW-MSA. Each CNN layer includes two 3x3 convolutions, which take the results extracted by the Transformer and feed them into the convolutions to further extract local features. S3. Add feature maps of different scales to the multi-scale skip connection module to extract feature information at multiple scales; In step S3, the multi-scale skip connection module includes several convolutional kernels of different sizes. The feature maps of each stage are input into convolutions with kernels of different sizes to obtain feature maps of different scales. The feature maps of different scales are concatenated and stitched together according to the channel dimension to obtain the output of the multi-scale skip connection module, which includes: S31. Combine feature maps of different levels Each operation undergoes several parallel convolutional operations: This represents a convolution with a kernel size of k; S32. Convert the convolution result Cascaded fusion along the channel dimension yields a feature map containing multi-scale information. : This indicates cascading and splicing according to channel dimensions; S4. Fuse the feature information from multiple scales, and multiply the fused feature with the features of each scale by weights to obtain the final multi-scale adaptive receptive field selection result feature map. In step S4, for the obtained feature map Processing it with a 1x1 convolution kernel yields the same dimension. Then cascaded fusion along the channel dimension to obtain feature maps containing multi-scale information. : feature map The input is fed into the compressed pyramid gating unit module, and after three sequential large kernel selection convolutions, three features of three adaptive receptive fields are obtained respectively. And they are spliced ​​together to obtain the feature map. : Then, It is reconstructed into a matrix through 1x1 convolution transpose and reshaping operations. sum matrix : exist and Perform matrix multiplication and then reshape the matrix to obtain the eigenvectors. : Then use 1×1 convolution and reshaping operations to map the features. Mapped to a matrix : and Perform matrix multiplication between the two channels, followed by a reshaping operation to obtain the channel attention weight vector. : Perform 1×1 convolutions between inputs, layer normalization, and apply the sigmoid activation function to obtain the weights. : With fusion features Element-wise multiplication yields the final feature map of the multi-scale adaptive receptive field selection result. : In the above process, reshape represents the reshaping operation. This represents the sigmoid activation function. Representation layer normalization, Represents matrix multiplication. This indicates element-wise multiplication; S5. The above process is trained using cross-entropy loss, and the trained data is then used for water body extraction from remote sensing images.

2. The method for extracting water bodies from remote sensing images based on multi-scale feature fusion using compressed pyramid gating units according to claim 1, characterized in that: In step S1, the preprocessing of the dataset includes cropping, labeling, and binarizing the label pixels of the labeled images; The dataset partitioning includes dividing the dataset into training, validation, and test sets according to a preset ratio; Data augmentation of the dataset includes: random scaling, random horizontal and vertical flipping, random Gaussian noise, contrast-limited adaptive histogram equalization, random rotation, and random cropping.

3. The method for extracting water bodies from remote sensing images based on multi-scale feature fusion using compressed pyramid gating units according to claim 1, characterized in that: In step S5, the AdamW optimizer and poly learning rate adjustment strategy are used to calculate the loss between the predicted and ground truth maps using the cross-entropy loss function. The cross-entropy loss calculation formula is as follows: Where N represents the total number of pixels in the image, ∈{0,1} represents a pixel The true label, Indicates for pixels The probability of belonging to category 1.

Images