Remote sensing image segmentation method combining positioning extraction and pulse channel enhancement

Abstract

The invention discloses a remote sensing image segmentation method combining positioning extraction and pulse channel enhancement, and relates to the field of remote sensing image segmentation, and the method comprises the following steps: extracting the space and position information of a remote sensing image through employing a positioning feature extraction network, and generating a first enhanced feature map fusing multi-level features; based on the first enhanced feature map, performing multi-scale feature enhancement by using a pulse channel feature enhancement network to generate a second enhanced feature map; and based on the second enhanced feature map, performing category prediction by using the remote sensing image segmentation network, and outputting a segmentation result of the remote sensing image. Through positioning feature extraction, pulse channel enhancement and compound loss optimization, the method can be suitable for high-resolution remote sensing image processing and resource-constrained environments, and the precision and efficiency of remote sensing image segmentation are significantly improved.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing image segmentation, and more specifically, to a remote sensing image segmentation method combining location extraction and pulse channel enhancement. Background Art

[0002] Remote sensing image segmentation is a core task in remote sensing technology and is widely used in fields such as environmental monitoring, urban planning, agricultural resource management, and disaster assessment. With the development of remote sensing technology, high-resolution images provide rich details of ground objects, but at the same time, they also face challenges such as complex scenes, multi-scale distribution of targets, and dynamic interference, which pose higher requirements for the accuracy and efficiency of segmentation algorithms.

[0003] Traditional remote sensing image segmentation methods mostly rely on manually designed features or shallow convolutional networks, which are difficult to fully extract the fine spatial information in high-resolution images, especially in complex scenes, where edge details are easily lost. In addition, existing methods have poor adaptability to multi-scale targets and cannot effectively capture the features of ground objects of different sizes. For example, there are significant differences in feature expressions between small-scale targets and large-scale targets, resulting in limited segmentation accuracy.

[0004] In addition, advanced methods based on deep learning, including deep convolutional networks and Transformer models, although they improve the segmentation accuracy, their model complexity is high, and the training and inference processes consume a large amount of computing resources, making it difficult to be deployed to edge devices or real-time processing scenarios. At the same time, the existing methods have limited ability to perceive dynamic changes and lack an efficient channel feature enhancement mechanism, resulting in insufficient robustness of the segmentation results under dynamic interference.

[0005] Therefore, how to design a remote sensing image segmentation method combining location extraction and pulse channel enhancement to effectively solve the defects of low segmentation accuracy, high resource consumption, and insufficient dynamic response in the prior art is an urgent problem for those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides a remote sensing image segmentation method combining location extraction and pulse channel enhancement, which realizes the accurate capture and enhancement of multi-scale features and significantly improves the segmentation accuracy. At the same time, it greatly reduces the demand for computing resources, enabling the method to operate efficiently in an environment with limited resources while maintaining high accuracy, providing a flexible and powerful solution for remote sensing image segmentation in complex scenes.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] The present invention provides a remote sensing image segmentation method combining location extraction and pulse channel enhancement, including the following steps:

[0009] S1. Use the positioning feature extraction network to extract the spatial and location information of the remote sensing image and generate the first enhanced feature map that fuses multi-level features;

[0010] S2. Based on the first enhanced feature map, use the pulse channel feature enhancement network to perform multi-scale feature enhancement and generate the second enhanced feature map;

[0011] S3. Based on the second enhanced feature map, use the remote sensing image segmentation network to perform class prediction and output the segmentation result of the remote sensing image.

[0012] Preferably, the S1 includes:

[0013] S11. Input the remote sensing image into the Mamba pre-trained model to extract the initial high-level semantic features

[0014] S12. Perform low-level feature extraction and stabilization processing on the initial high-level semantic features through the convolutional layer and the normalization layer to obtain the normalized features to obtain the normalized features

[0015] S13. Inject the position encoding, perform full-dimensional dynamic convolution and linear projection on the normalized features in sequence to generate the intermediate features α, β, γ containing spatial perception; to generate the intermediate features α, β, γ containing spatial perception;

[0016] S14. Use spatial relationship modeling and skip connection fusion to fuse the intermediate features α, β, γ with the normalized features to obtain the fused features

[0017] S15. Extract complex spatial features from the fused features through the convolutional layer and the windmill-shaped convolutional layer, and enhance the sensitivity of key positions through the position trigger; to enhance the sensitivity of key positions through the position trigger;

[0018] S16. Perform non-linear transformation on the processing results of the windmill-shaped convolutional layer and the position trigger through the ReLU activation function and perform fusion to obtain the first enhanced feature map

[0019] Preferably, the S14 includes:

[0020] Transpose and reshape the intermediate features α and β respectively and then multiply them to obtain the spatial correlation matrix, and normalize the spatial correlation matrix through the softmax activation function to obtain the normalized spatial correlation matrix;

[0021] Multiply the normalized spatial correlation matrix by the reshaped intermediate feature γ to obtain the feature with enhanced spatial information

[0022] The features after enhancing the spatial information are reshaped and fused with the normalized features to obtain the fused features

[0023] Preferably, the position trigger includes: two convolutional layers, an activation function, and a fully connected layer.

[0024] Preferably, the S2 includes:

[0025] S21. Input the first enhanced feature map into the first spiking neuron layer, the second spiking neuron layer, and the convolutional layer respectively, and perform reshaping to obtain features δ, ∈, and ζ;

[0026] S22. Based on the features δ, ∈, and ζ, perform spiking feature weighting and spatial relationship modeling to obtain a feature map

[0027] S23. Perform skip connection fusion on the feature map and the first enhanced feature map to obtain a feature map

[0028] S24. Enhance the channel response ability and expand the receptive field of the feature map through a channel trigger and a dilated convolutional layer, and apply a depthwise separable convolutional layer for depthwise separable convolution;

[0029] S25. Fuse the processing results of the dilated convolutional layer and the depthwise separable convolutional layer to obtain the second enhanced feature map

[0030] Preferably, the S22 includes:

[0031] Multiply the features ∈ and ζ and normalize them through the Softmax activation function to obtain the relationship weights between channels;

[0032] Multiply the relationship weights with the feature δ and obtain a feature map through reshaping

[0033] Preferably, the channel trigger includes: a global average pooling layer, a fully connected layer, a convolutional layer, and an activation function.

[0034] Preferably, in the S3, using a remote sensing image segmentation network for class prediction includes:

[0035] Input the second enhanced feature map into a convolutional layer and an upsampling layer to obtain a class prediction map

[0036] Perform class assignment on the class prediction map through the Softmax activation function to obtain the probability distribution of each pixel point and generate a prediction probability map.

[0037] Preferably, the remote sensing image segmentation network adopts a composite loss function; the composite loss function is a weighted combination of the cross-entropy loss function and the Dice loss function, which is used to balance the pixel-level classification accuracy and the segmentation region overlap degree.

[0038] Preferably, in the S3, it further includes:

[0039] Perform thresholding on the prediction probability map output by the remote sensing image segmentation network to obtain the corresponding binary segmentation result.

[0040] It can be seen from the above technical solutions that compared with the prior art, the technical solutions of the present invention have the following

[0041] Beneficial effects:

[0042] 1. The location feature extraction network extracts initial high-level semantic features through a pre-trained model, combines dynamic convolution, position encoding, and spatial relationship modeling, can dynamically adjust the convolution kernel size to adapt to different scale targets, and retains low-level details and high-level semantic information through skip connection fusion. It significantly improves the model's perception ability for complex scenes, solves the problem of insufficient accuracy in multi-scale target segmentation, and is especially suitable for the refined processing requirements of high-resolution remote sensing images.

[0043] 2. The pulse channel feature enhancement network enhances the response ability to dynamic changes by introducing a pulse neuron layer, and uses dilated convolution to expand the receptive field to capture long-range spatial dependencies. Combining depthwise separable convolution to optimize the calculation efficiency, it can significantly reduce the calculation complexity while ensuring the feature enhancement effect. It can adapt to the real-time processing requirements of edge computing devices and meet the efficient deployment in low-storage environments and resource-constrained scenarios.

[0044] 3. In the remote sensing image segmentation network, the class imbalance problem is optimized through the weighted combination of cross-entropy loss and Dice loss, and the probability map is converted into a binary segmentation mask by combining an adaptive threshold adjustment strategy. It effectively alleviates the challenge of uneven distribution of ground object classes in remote sensing images, and at the same time improves the robustness and practicality of the segmentation results. Description of the Drawings

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on the provided drawings.

[0046] Figure 1 Schematic diagram of a remote sensing image segmentation process combining location extraction and pulse channel enhancement provided by an embodiment of the present invention;

[0047] Figure 2 Schematic diagram of the process of generating a first enhanced feature map using a location feature extraction network provided by an embodiment of the present invention;

[0048] Figure 3 Structure framework diagram of a position trigger provided by an embodiment of the present invention;

[0049] Figure 4 Schematic diagram of the process of generating a second enhanced feature map using a pulse channel feature enhancement network provided by an embodiment of the present invention;

[0050] Figure 5 Structure framework diagram of a channel trigger provided by an embodiment of the present invention;

[0051] Figure 6 Flowchart of a remote sensing image segmentation method in the agricultural land classification scenario provided by an embodiment of the present invention. Detailed implementation manners

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0053] Embodiment 1;

[0054] As Figure 1 shown, this embodiment provides a remote sensing image segmentation method combining location extraction and pulse channel enhancement, including the following steps:

[0055] S1. Use a location feature extraction network to extract the spatial and location information of the remote sensing image and generate a first enhanced feature map that fuses multi-level features;

[0056] S2. Based on the first enhanced feature map, use a pulse channel feature enhancement network for multi-scale feature enhancement to generate a second enhanced feature map;

[0057] S3. Based on the second enhanced feature map, use a remote sensing image segmentation network to perform class prediction and output the segmentation result of the remote sensing image.

[0058] By integrating the dynamic convolution, position encoding, and spatial relationship modeling of the location feature extraction network, this method improves the model's perception ability of complex scenes and solves the problem of insufficient multi-scale object segmentation accuracy; the pulse channel feature enhancement network optimizes the feature representation through the pulse neuron layer and dilated convolution, achieving efficient calculation and real-time processing; while the remote sensing image segmentation network uses a composite loss function and adaptive threshold adjustment, effectively alleviating the class imbalance problem and enhancing the robustness and practicality of the segmentation result. It is particularly suitable for the refined processing of high-resolution remote sensing images and efficient deployment in resource-constrained environments.

[0059] The following further elaborates on each step in the above method:

[0060] As Figure 2 shown, in step S1 of this embodiment, use a location feature extraction network to extract the spatial and position information of the remote sensing image and generate a first enhanced feature map that fuses multi-level features; specifically including:

[0061] S11. Input the remote sensing image into the Mamba pre-trained model to extract initial high-level semantic features The pre-training of this model on remote sensing images enables it to fully capture global and local structural information, providing a highly robust semantic basis for subsequent operations;

[0062] S12. Perform low-level feature extraction and stabilization processing on the initial high-level semantic features through a convolutional layer and a normalization layer to obtain the normalized features

[0063] S13. Inject position encoding, perform full-dimensional dynamic convolution, and linear projection on the normalized features in sequence to generate intermediate features α, β, γ that contain spatial perception;

[0064] S14. Use spatial relationship modeling and skip connection fusion to fuse the intermediate features α, β, γ with the normalized features to obtain the fused features

[0065] S15. Perform Extract complex spatial features through convolutional layers and windmill-shaped convolutional layers, and enhance the sensitivity to key positions through position triggers; it introduces a windmill-shaped convolutional layer and a position trigger. The former extracts complex spatial features through multi-directional convolutional kernels, and the latter enhances the sensitivity of the model to edges and object details by weighting key regions through activation functions, thus achieving more accurate spatial feature modeling in complex scenarios;

[0066] S16. Non-linearly transform the processing results of the windmill-shaped convolutional layer and the position trigger through the ReLU activation function, and fuse them to obtain the first enhanced feature map

[0067] It should be noted that the convolutional layer in this embodiment is a standard convolutional layer, which uses a convolutional kernel with a fixed size of 3×3 or 5×5, and the stride Stride is 1 to extract local spatial features.

[0068] Further S14 includes:

[0069] Transpose and reshape the intermediate features α and β respectively, then multiply them to obtain a spatial correlation matrix, and normalize the spatial correlation matrix through the softmax activation function to obtain a normalized spatial correlation matrix;

[0070] Multiply the normalized spatial correlation matrix by the reshaped intermediate feature γ to obtain a feature with enhanced spatial information

[0071] For the feature with enhanced spatial information Reshape it and fuse it with the normalized feature To obtain a fused feature

[0072] It models the position relationship through the spatial correlation matrix, and uses transpose, reshape and Softmax normalization operations to explicitly encode the dynamic spatial dependence relationships of the intermediate features α, β, and γ, solving the difficulty of modeling the correlation of distant targets in remote sensing images. When fusing the normalized features, skip connections are used, effectively retaining the detailed information of the initial features, avoiding the problem of gradient disappearance in the training of deep networks, and further improving the stability of feature fusion and the coherence of segmentation results.

[0073] As Figure 3 shown, the position trigger includes: two convolutional layers, an activation function and a fully connected layer. It extracts local position features through convolutional layers, and the fully connected layer learns the global weight distribution, and combines the activation function to dynamically adjust the response intensity of key positions. It enhances the model's ability to locate sparsely distributed targets in remote sensing images.

[0074] As Figure 4As shown, in this embodiment S2, based on the first enhanced feature map, a multi-scale feature enhancement is performed using a pulse channel feature enhancement network to generate a second enhanced feature map; specifically including:

[0075] S21. Input the first enhanced feature map into the first pulse neuron layer, the second pulse neuron layer, and the convolutional layer respectively, and perform reshaping to obtain features δ, ∈, ζ; it processes the first enhanced feature map through the pulse neuron layer, utilizes the timing characteristics of the pulse signal to enhance the adaptability of the model to dynamic changes, and solves the problem of insufficient response of the traditional convolutional network to time series features.

[0076] S22. Based on the features δ, ∈, ζ, perform pulse feature weighting and spatial relationship modeling to obtain a feature map

[0077] S23. Perform skip connection fusion on the feature map and the first enhanced feature map to obtain a feature map

[0078] S24. For the feature map enhance the channel response ability and expand the receptive field through a channel trigger and a dilated convolutional layer, and apply a depthwise separable convolutional layer for depthwise separable convolution; it combines dilated convolution and depthwise separable convolution. The former captures long-range dependencies by expanding the receptive field, and the latter reduces the computational amount by decomposing the convolution operation. While ensuring the multi-scale feature extraction ability, it significantly reduces resource consumption.

[0079] S25. Fuse the processing results of the dilated convolutional layer and the depthwise separable convolutional layer to obtain the second enhanced feature map

[0080] Further, S22 includes: multiplying the features ∈, ζ and normalizing through the Softmax activation function to obtain the relationship weights between channels; multiplying the relationship weights with the feature δ and performing reshaping to obtain a feature map

[0081] It realizes the adaptive enhancement of key channels by normalizing the relationship weights between channels through Softmax and combining the weighted fusion of the feature δ, aiming to highlight the discriminative features of different ground object categories in the remote sensing image, thereby improving the category discrimination of the segmentation result in complex scenes.

[0082] Such as Figure 5As shown, the channel trigger includes: a global average pooling layer, a fully connected layer, a convolutional layer, and an activation function. The channel trigger contains a global average pooling layer and a fully connected layer. The former compresses global spatial information to generate a channel descriptor, and the latter learns the non-linear relationship between channels, combines with the convolutional layer to dynamically adjust channel weights, effectively suppressing the interference of redundant channels and strengthening the high-frequency features of the model in remote sensing images.

[0083] In this embodiment S3, based on the second enhanced feature map, a remote sensing image segmentation network is used for class prediction, and the segmentation result of the remote sensing image is output.

[0084] In this step, first, using the remote sensing image segmentation network for class prediction includes:

[0085] Input the second enhanced feature map into the convolutional layer and the upsampling layer to obtain a class prediction map Through the Softmax activation function for the class prediction map Perform class assignment to obtain the probability distribution of each pixel point And generate a predicted probability map. It restores the resolution of the feature map through the upsampling layer, combines with the Softmax activation to generate a pixel-level probability distribution, ensuring that the segmentation result is consistent with the size of the original image.

[0086] Furthermore, the remote sensing image segmentation network adopts a composite loss function; the composite loss function is a weighted combination of the cross-entropy loss function and the Dice loss function, which is used to balance the pixel-level classification accuracy and the segmentation region overlap degree. In the composite loss function, the cross-entropy loss is used to optimize the pixel-level classification accuracy, and the weighted combination of the Dice loss is used to measure and improve the region consistency through the overlap degree, solving the model deviation problem caused by class imbalance in remote sensing images. Combining with dynamically adjusting the weight parameters, it can flexibly adapt to the needs of different datasets and enhance the generalization ability of the model.

[0087] Furthermore, threshold processing is performed on the predicted probability map output by the remote sensing image segmentation network to obtain the corresponding binary segmentation result.

[0088] This embodiment proposes a remote sensing image segmentation method combining location extraction and pulse channel enhancement. The spatial and position information of the remote sensing image is captured by the location feature extraction network, multi-scale feature enhancement is performed using the pulse channel feature enhancement network, and finally the segmentation result is output through the remote sensing image segmentation network. This method integrates advanced technologies such as dynamic convolution, position encoding, pulse neuron layer, and composite loss function, effectively improving the accuracy and efficiency of remote sensing image segmentation, and is particularly suitable for the refined processing of high-resolution images and the efficient deployment in resource-constrained environments.

[0089] Embodiment 2;

[0090] In this embodiment, it specifically relates to the application of a remote sensing image segmentation method based on positioning extraction and pulse channel enhancement in the scenario of agricultural land classification;

[0091] Agricultural land classification is an important task for precision agricultural management, crop monitoring, and land resource assessment. High-resolution remote sensing images can provide details such as farmland boundaries, crop types, and irrigation facilities. However, due to the large scale differences in farmland areas (such as small plot farmland and large-scale planting areas), complex crop textures, and seasonal variations, traditional segmentation methods are vulnerable to dynamic interference, resulting in blurred classification boundaries and class confusion.

[0092] As Figure 6 shown, the specific steps of this method include:

[0093] 1) Preprocessing of agricultural remote sensing images and extraction of positioning features;

[0094] First, input the multi-spectral remote sensing image, perform radiometric correction, geometric registration, and normalization processing, retain the bands sensitive to vegetation such as near-infrared and red edge, and enhance the distinguishability of crop features. Then, input the processed image into the Mamba pre-trained model to extract the global farmland distribution and local plot boundary features.

[0095] For the agricultural scenario, prior information of farmland grids is introduced into the position encoding here to enhance the model's perception ability of regular farmland boundaries; and multi-scale convolutional kernels are set in the full-dimensional dynamic convolution to capture the texture of small plot farmland and the spatial continuity features of large-scale planting areas respectively.

[0096] 2) Pulse channel feature enhancement and multi-scale fusion;

[0097] In this step, according to the growth time series characteristics of crops, a time decay factor is introduced into the pulse neuron layer to enhance the model's dynamic response ability to seasonal changes. Further, the weights of the vegetation index-related channels are strengthened through channel triggers to suppress the interference in cloud-covered or shadow areas.

[0098] In addition, depthwise separable convolution is used to replace the standard convolution, which reduces the computational amount while ensuring multi-scale feature extraction and adapts to the deployment requirements of UAV edge devices.

[0099] 3) Agricultural land classification and post-processing;

[0100] In the remote sensing image segmentation network, adaptive weights are set for agricultural scenario categories (such as wheat, corn, fallow land, water body) to balance the problem of sample imbalance. A boundary loss term is added to the composite loss function to strengthen the segmentation accuracy of farmland boundaries.

[0101] Further, perform morphological closing operation on the output probability map to fill small holes and smooth the boundaries; combine the prior knowledge of the farmland vector map to correct the abnormal areas in the segmentation result.

[0102] The remote sensing image segmentation method based on positioning extraction and pulse channel enhancement in this embodiment can be applied to agricultural resource census, crop growth monitoring and precise irrigation planning, and is especially suitable for realizing high-precision and high-efficiency farmland plot segmentation and classification tasks in multi-spectral remote sensing images.

[0103] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other. For the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0104] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but will be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A remote sensing image segmentation method combining positioning extraction and pulse channel enhancement, characterized in that: The following steps are involved: S1, using the positioning feature extraction network to extract the spatial and position information of the remote sensing image, and generate a first enhanced feature map that integrates multi-level features; S2. Based on the first enhanced feature map, a pulse channel feature enhancement network is used to perform multi-scale feature enhancement to generate a second enhanced feature map; S3. Based on the second enhanced feature map, use the remote sensing image segmentation network to perform category prediction and output the segmentation result of the remote sensing image.

2. A remote sensing image segmentation method combining positioning extraction and pulse channel enhancement according to claim 1, characterized in that: Said S1 comprises: S11. Input remote sensing images into the Mamba pre-trained model to extract initial high-level semantic features S12, initial high-level semantic features through convolutional layers and normalization layers Perform low-level feature extraction and stabilization to obtain normalized features S13. Normalized features Sequentially inject position encoding, perform full-dimensional dynamic convolution and linear projection to generate intermediate features α, β, γ containing spatial perception; S14, using spatial relationship modeling and skip link fusion, the intermediate features α, β, γ and the normalized features Fusion is performed to obtain the fused features S15. Features after fusion Complex spatial features are extracted through convolutional layers and pinwheel convolutional layers, and key position sensitivity is enhanced through position triggers; S16. Perform nonlinear transformation on the windmill convolution layer and the position trigger processing results through the ReLU activation function, and fuse them to obtain the first enhanced feature map.

3. A remote sensing image segmentation method combining positioning extraction and pulse channel enhancement according to claim 2, characterized in that: The S14 comprises: The intermediate features α and β are transposed and reshaped respectively and then multiplied to obtain a spatial correlation matrix, and the spatial correlation matrix is ​​normalized by a softmax activation function to obtain a normalized spatial correlation matrix; Multiply the normalized spatial correlation matrix with the reshaped intermediate feature γ to obtain the feature after spatial information enhancement Features after spatial information enhancement Reshape and normalize the features Fusion is performed to obtain the fused features 4. The remote sensing image segmentation method combining positioning extraction and pulse channel enhancement according to claim 2, characterized in that: The position trigger includes: two convolutional layers, an activation function and a fully connected layer.

5. The remote sensing image segmentation method combining positioning extraction and pulse channel enhancement according to claim 1, characterized in that: The S2 comprises: S21, the first enhanced feature map Input the first pulse neuron layer, the second pulse neuron layer and the convolution layer respectively, and reshape them to obtain features δ, ∈, ζ; S22. Based on the features δ, ∈, ζ, pulse feature weighting and spatial relationship modeling are performed to obtain a feature map S23, feature map With the first enhanced feature map Perform skip link fusion to obtain feature maps S24. Feature map Enhance channel responsiveness and expand receptive field through channel triggers and dilated convolution layers, and apply depthwise separable convolution layers for depthwise separable convolution; S25, fusing the processing results of the dilated convolution layer and the depth-separable convolution layer to obtain a second enhanced feature map 6. A remote sensing image segmentation method combining positioning extraction and pulse channel enhancement according to claim 5, characterized in that: The S22 comprises: Multiply the features ∈ and ζ and normalize them through the Softmax activation function to obtain the relationship weights between channels; Multiply the relationship weight by the feature δ and reshape to obtain the feature map 7. The remote sensing image segmentation method combining positioning extraction and pulse channel enhancement according to claim 5, characterized in that: The channel trigger includes: a global average pooling layer, a fully connected layer, a convolutional layer and an activation function.

8. The remote sensing image segmentation method combining positioning extraction and pulse channel enhancement according to claim 1, characterized in that: In S3, the remote sensing image segmentation network is used to perform category prediction, including: The second enhanced feature map is input into the convolution layer and the upsampling layer to obtain the category prediction map The category prediction graph is obtained by using the Softmax activation function Perform category assignment to obtain the probability distribution of each pixel And generate a predicted probability map.

9. The remote sensing image segmentation method combining positioning extraction and pulse channel enhancement according to claim 1, characterized in that: The remote sensing image segmentation network adopts a composite loss function; the composite loss function is a weighted combination of a cross entropy loss function and a Dice loss function, which is used to balance pixel-level classification accuracy and segmentation area overlap.

10. The remote sensing image segmentation method combining positioning extraction and pulse channel enhancement according to claim 1, characterized in that: The S3 further includes: The predicted probability map output by the remote sensing image segmentation network is thresholded to obtain the corresponding binary segmentation result.

Images