Land development category segmentation method and system based on edge detection of remote sensing image

Abstract

The application relates to the technical field of satellite remote sensing image segmentation, in particular to a remote sensing image land development category segmentation method and system based on edge detection, and specifically as follows: a satellite remote sensing image land development category segmentation model is constructed, satellite remote sensing land images captured by satellite remote sensing are input into the model, and a land development category segmentation result is obtained; the model comprises a satellite image feature reconstruction module, a super-resolution satellite image detail enhancement module, a multi-scale convolution construction super-resolution satellite image detail enhancement module, an edge local perception attention module, a satellite image global capture module and a satellite image land category segmentation module. The application can solve the problems of fuzzy ground object boundary, difficult multi-scale feature fusion and difficult local details and global semantic information collaborative modeling when processing satellite remote sensing land images, and effectively improves the segmentation precision of satellite remote sensing images on the development land category.

Description

Technical Field

[0001] This invention relates to the field of satellite remote sensing image segmentation technology, and in particular to a method and system for segmenting land development categories in remote sensing images based on edge detection. Background Technology

[0002] With the development of deep learning and high-resolution satellite remote sensing technology, land classification using satellite remote sensing image segmentation technology has become an important method for smart city construction and natural resource management. Satellite remote sensing images can accurately segment dozens of land types, including construction land, arable land, forest land, and water areas. Its applications cover multiple dimensions, including land spatial planning, crop planting monitoring, ecological environment assessment, and disaster loss quantification. Satellite remote sensing image land segmentation technology can dynamically monitor macro trends such as urban expansion, conversion of arable land to non-grain crops, and changes in forest cover, assisting government departments in formulating strategies for optimal allocation of land resources. Compared to traditional manual segmentation methods, deep learning-based satellite remote sensing land segmentation methods can improve classification efficiency by nearly 100 times, significantly reducing the cost of high-precision land surveys and providing timely and spatially granular data support for global climate change research and the implementation of sustainable development goals.

[0003] Despite significant progress in deep learning-based satellite remote sensing land segmentation technology, its application still faces many challenges: First, complex land features place higher demands on the model's feature extraction capabilities, and existing network architectures often suffer from blurred edges or missed detections of small targets; second, dynamic environmental changes can significantly reduce the model's generalization performance, requiring frequent data retraining to maintain accuracy.

[0004] Therefore, this invention proposes a method and system for segmenting land development categories in remote sensing images based on edge detection to solve the above problems. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention develops a method and system for land development category segmentation in remote sensing images based on edge detection. By solving problems such as blurred land feature boundaries, difficulties in multi-scale feature fusion, and the inability to coordinate the modeling of local details and global semantic information when processing satellite remote sensing land images, this invention can effectively improve the segmentation accuracy of land categories in satellite remote sensing images.

[0006] On the one hand, the technical solution of this invention to solve the technical problem is a land development category segmentation method based on edge detection in remote sensing images. This method constructs a land development category segmentation model based on satellite remote sensing images, inputs satellite remote sensing land images into the model, and obtains the land development category segmentation result. The method includes the following steps:

[0007] S1. Construct a satellite image feature reconstruction module: This module includes depthwise separable convolution operations. The activation function, satellite image feature space reconstruction module, and downsampling layer input satellite remote sensing land images captured by satellite remote sensing into the satellite image feature reconstruction module and output satellite image feature reconstruction features.

[0008] S2. Construct a super-resolution satellite image detail enhancement module: Construct a super-resolution satellite image detail enhancement module based on multi-scale convolution. Input the satellite image feature reconstruction features into the super-resolution satellite image detail enhancement module to perform super-resolution detail enhancement, and output the super-resolution satellite image detail enhancement features.

[0009] S3. Construct an edge local perception attention module: This module includes an edge local perception attention module, which inputs the super-resolution satellite image detail enhancement features into the edge local perception attention module for local feature capture and outputs edge local perception attention features.

[0010] S4. Construct a global satellite image capture module: This module includes a global satellite image capture module, which inputs edge local perception attention features into the global satellite image capture module to enhance the global correlation of features and outputs global satellite image features;

[0011] S5. Construct a satellite image land category segmentation module: Combine local edge awareness attention features with global satellite image features through feature concatenation to output fused local and global information features. Then, input these features into the satellite image land category segmentation module to output a satellite image land category segmented image. .

[0012] S1 is as follows:

[0013] S1.1 Satellite remote sensing land images The kernel size input to the satellite image feature reconstruction module is Depth separable convolutional layers and Activation function, output features ;

[0014] S1.2, Features The satellite image feature space reconstruction module, which inputs the data into the satellite image feature reconstruction module, first modifies the features through a transpose operation. The spatial structure, after transpose operation, outputs transformed features. Transformation characteristics After passing through a convolution kernel with a size of The convolutional layer is then processed by feature segmentation. Operations will transform features The number of channels segmented along the channel direction is a transformation feature. sub-features and ;

[0015] (1) For sub-features This feature is processed by a convolution kernel with a size of Depth-separable convolution operations Nonlinear activation function operations Feature reconstruction operation Feature scanning operation Normalization operation, and reuse Reconstruction operation, outputting the processed sub-features The first time I passed by The feature reconstruction operation flattens the features into a one-dimensional vector, and the second pass... Feature reconstruction operations restore features to two-dimensional vectors;

[0016] (2) For sub-features After convolution kernel size is The convolutional layer and kernel size are Depth-separable convolutional layers and Nonlinear activation function operation, outputting processed sub-features ;

[0017] (3) Process the sub-features and The components are spliced ​​and merged, and the kernel size is [missing information]. The convolutional layer outputs features Then the features With features splicing and fusion to output features ;

[0018] S1.3, Features The input is fed into a downsampling layer for downsampling, and the output is satellite image feature reconstruction features. .

[0019] S2 is as follows:

[0020] The super-resolution satellite image detail enhancement module based on multi-scale convolution specifically includes three parallel convolutional layers that reconstruct features from satellite image features. Input into the super-resolution satellite image detail enhancement module;

[0021] S2.1 Reconstructing features from satellite image features The inputs are fed into three parallel convolutional layers respectively;

[0022] For the convolutional layer of the first branch: features After convolution kernel size is Depth-separable convolutional layers output features ;feature After passing through a convolution kernel with a size of Depth-separable convolutional layers output features ;

[0023] For the convolutional layer in the second branch: features After convolution kernel size is Convolutional layers and Nonlinear activation function, output features ;

[0024] For the third branch of the convolutional layer: features After convolution kernel size is The depth of the separable convolutional layer yields features and features ,feature splicing and fusion to output features ;feature After passing through a convolution kernel with a size of Depth-separable convolutional layers output features ;

[0025] The features output by the three branches of the convolutional layer ,feature and characteristics Perform splicing and fusion to output features By stitching features together using multi-scale convolutional layers, the multi-scale representation capability of detailed features in super-resolution satellite images can be enhanced.

[0026] S2.2, Features After passing through a global average pooling layer and a fully connected layer, it is then combined with features. Fusion, output channel attention features Then the features After convolution kernel size is After the convolutional layer and features Fusion, output spatial attention features Then, channel attention features Spatial attention characteristics The components are spliced ​​and merged, and then processed by a convolution kernel with a size of [missing value]. The convolutional layers output super-resolution satellite image detail enhancement features. .

[0027] S3 is as follows:

[0028] Enhance details in super-resolution satellite images The input is fed into the edge local perception attention module, which consists of two branches, which process the features... The two branches of the edge local perception attention module are respectively input to extract local channel features and local spatial features. Then, the two features are concatenated and fused to obtain the edge local perception attention features. ;

[0029] S3.1 Enhance details in super-resolution satellite images The features are input into the first branch of the edge local awareness attention module. The convolutional layers are sequentially processed through a global average pooling layer with a kernel size of [missing value]. convolutional layers and Activation function, output features , will feature With features Perform broadcast multiplication, then multiply the output feature with the feature... Feature concatenation is performed to ultimately output local channel features. ;

[0030] S3.2 Enhance details in super-resolution satellite images The input is fed into the second branch of the edge local awareness attention module, and the features... After convolution kernel size is The convolutional layer outputs features Then the features The input is fed into three parallel depthwise separable convolutional layers, each with a kernel size of [missing value]. The depth of the separable convolutional layer and the kernel size are The depth of the separable convolutional layer and the size of the convolutional kernel are The depthwise separable convolutional layers are used, and then the output features of the three parallel depthwise separable convolutional layers are concatenated to output features. Then the features After convolution kernel size is convolutional layers and Activation function, output features Finally, the features With features Perform broadcast multiplication, then multiply the output feature with the feature... Feature concatenation is performed to ultimately output local spatial features. ;

[0031] S3.3, Local channel features With local spatial features By splicing and fusion, the edge-local awareness attention features are output. .

[0032] The S4 steps are as follows:

[0033] Enhance details in super-resolution satellite images The data is input into the satellite image global capture module, which includes two branches, to process the features. The two branches of the satellite image global capture module are input separately to obtain the first branch features and the second branch features. Then, the two branch features are compared with the feature... By stitching and fusion, global features of the satellite image are obtained. ;

[0034] S4.1 Enhance details in super-resolution satellite images The data is input into the first branch of the satellite image global capture module, where features are first processed. Perform a dimension transpose operation to convert the features Rotate along the height dimension to output features Then the features The input to the convolution kernel size is The depth can be separated into convolutional layers and batch normalization layers, and the output features are then compared with the feature... Perform feature concatenation and output features. ,feature After passing through The activation function operation and the convolution kernel size are The convolutional layer outputs features ;

[0035] Features After convolution kernel size is The depth can be separated into convolutional layers and batch normalization layers, which will combine the output features with the feature set. Perform feature concatenation and output features. , will feature After passing through a convolution kernel with a size of The depth can be separable into convolutional layers, batch normalized layers, and convolutional kernels with a size of [missing value]. The convolutional layer outputs the first branch features. ;

[0036] S4.2 Enhance details in super-resolution satellite images The data is input into the second branch of the satellite image global capture module, where features are first processed. Perform a dimension transpose operation to convert the features Rotate along the width dimension to output features Then the features The input to the convolution kernel size is The depth can be separated into convolutional layers and batch normalization layers, and the obtained output is combined with the enhanced features. Perform feature concatenation and output features. Then the features go through Activation function operations and convolution kernel size are The convolutional layer outputs features ;

[0037] Enhance features After convolution kernel size is The depth can be separated into convolutional layers and batch normalization layers, which will combine the output features with the feature set. Perform feature concatenation and output features. ; Features After convolution kernel size is The depth can be separable into convolutional layers, batch normalized layers, and convolutional kernels with a size of [missing value]. The convolutional layer outputs the second branch features. ;

[0038] S4.3, the first branch feature Second branch features With features By stitching and fusion, global features of satellite images are output. .

[0039] S5 is detailed below:

[0040] Edge local perception attention features Global features of satellite imagery Perform splicing and fusion to output fused local and global information features. Then the features The data is input into the satellite image land category segmentation module, and after passing through a convolution kernel with a size of [missing value], it is processed. The depth can separate convolutional layers and upsampling operations, and extract features. The resolution size is restored to match the image. The same applies to images after resolution restoration. The extreme point localization operation locates the specific land category to which each pixel in an image belongs, through... Color mapping operations map the specific land category to which a pixel belongs to a different color, generating a satellite image land category segmentation image. .

[0041] On the other hand, the present invention also provides a land development category segmentation system for remote sensing images based on edge detection, including a module for executing processing instructions for each step in a land development category segmentation method for remote sensing images based on edge detection.

[0042] The effects described in the invention are merely those of the embodiments, and not all the effects of the invention. The above technical solutions have the following advantages or beneficial effects:

[0043] This invention constructs a land development category segmentation model based on satellite remote sensing images. It processes the input satellite remote sensing land images through a combination of five modules. Specifically, through the synergistic effect of the satellite image feature reconstruction module and the super-resolution detail enhancement module, edge texture details can be enhanced while preserving global structural information, effectively solving the problems of blurred land feature boundaries and missed detection of small targets in traditional methods. Furthermore, by combining the dynamic capture of subtle features by the edge local perception attention module and the multi-dimensional transposed feature association modeling of the global capture module, a deep fusion of local details and large-scale land cover semantics is achieved, significantly improving the generalization ability of complex scenes.

[0044] This requires overcoming the collaborative conflicts caused by differences in feature dimensions and resolution between modules. Key information is dynamically weighted through a channel / spatial attention mechanism to eliminate redundancy. At the same time, the computational complexity is increased due to super-resolution multi-scale convolution and global dimension transpose operations. Specifically, a progressive architecture of "feature reconstruction → detail enhancement → local perception → global modeling → classification decision" is adopted to break through the limitations of the unidirectional information flow of traditional segmentation networks. An innovative inter-module interaction mechanism is designed to achieve hierarchical abstraction and feedback from low-level pixels to high-level semantics.

[0045] In summary, this invention discloses a method and system for land category segmentation in remote sensing images based on edge detection. It reconstructs effective information from land images using a satellite image feature reconstruction module; significantly enhances the representation of texture details at land cover boundaries in remote sensing images using a super-resolution satellite image detail enhancement module; captures subtle edge features in images using a local edge awareness attention module; and addresses the problem of semantic information inconsistency in large-scale land cover type classification using a global satellite image capture module. Ultimately, it effectively improves the accuracy of land category segmentation using satellite remote sensing images. Attached Figure Description

[0046] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0047] Figure 1 This is a schematic diagram of the overall structure of the method of the present invention.

[0048] Figure 2 This is a comparison of the segmentation results.

[0049] Figure 3 This is a structural diagram of the satellite image feature reconstruction module of the present invention.

[0050] Figure 4 This is a structural diagram of the super-resolution satellite image detail enhancement module of the present invention.

[0051] Figure 5This is a structural diagram of the edge local perception attention module of the present invention.

[0052] Figure 6 This is a structural diagram of the global satellite image capture module of the present invention.

[0053] Figure 7 This is a structural diagram of the satellite image land category segmentation module of the present invention. Detailed Implementation

[0054] To clearly illustrate the technical features of this solution, the invention will be described in detail below through specific implementation methods and in conjunction with the accompanying drawings.

[0055] Example 1

[0056] A land development category segmentation method based on edge detection in remote sensing images is proposed. This method constructs a land development category segmentation model based on satellite remote sensing images, inputting satellite-captured land images into the model to obtain the land development category segmentation results. The method includes the following steps:

[0057] S1. Construct a satellite image feature reconstruction module: This module includes depthwise separable convolution operations. The activation function, satellite image feature space reconstruction module, and downsampling layer input satellite remote sensing land images captured by satellite remote sensing into the satellite image feature reconstruction module and output satellite image feature reconstruction features.

[0058] S2. Construct a super-resolution satellite image detail enhancement module: Construct a super-resolution satellite image detail enhancement module based on multi-scale convolution. Input the satellite image feature reconstruction features into the super-resolution satellite image detail enhancement module to perform super-resolution detail enhancement, and output the super-resolution satellite image detail enhancement features.

[0059] S3. Construct an edge local perception attention module: This module includes an edge local perception attention module, which inputs the super-resolution satellite image detail enhancement features into the edge local perception attention module for local feature capture and outputs edge local perception attention features.

[0060] S4. Construct a global satellite image capture module: This module includes a global satellite image capture module, which inputs edge local perception attention features into the global satellite image capture module to enhance the global correlation of features and outputs global satellite image features;

[0061] S5. Construct a satellite image land category segmentation module: Combine local edge awareness attention features with global satellite image features through feature concatenation to output fused local and global information features. Then, input these features into the satellite image land category segmentation module to output a satellite image land category segmented image. .

[0062] In a specific implementation, S1 is as follows:

[0063] S1.1 Satellite remote sensing land images The kernel size input to the satellite image feature reconstruction module is Depth separable convolutional layers and Activation function, output features The calculation formula is:

[0064] ,

[0065] in, This indicates that the activation function performs non-linear processing on the features. Indicates the kernel size as The depth of the separable convolutional layer , Indicates the altitude of a satellite-sensed land image. Indicates the width of a satellite remote sensing land image. This indicates the number of channels in a satellite remote sensing land image.

[0066] S1.2, Features The satellite image feature space reconstruction module, which inputs the data into the satellite image feature reconstruction module, first modifies the features through a transpose operation. Spatial structure, by changing features The spatial structure facilitates subsequent reconstruction of feature information in different directions, and the transformed features are output after transpose operation. Transformation characteristics After passing through a convolution kernel with a size of The convolutional layer is then processed by feature segmentation. Operations will transform features The number of channels segmented along the channel direction is a transformation feature. sub-features and The calculation formula is:

[0067] ,

[0068] ,

[0069] in, This indicates the feature transpose operation. This indicates a feature segmentation operation. Indicates the kernel size as Convolutional layers;

[0070] (1) For sub-features This feature is processed by a convolution kernel with a size of Depth-separable convolution operations Nonlinear activation function operations Feature reconstruction operation Feature scanning operation Normalization operation, and reuse Reconstruction operation, outputting processed sub-features The first time I passed by The feature reconstruction operation flattens the features into a one-dimensional vector, and the second pass... Feature reconstruction restores features to two-dimensional vectors. The calculation formula is as follows:

[0071] ,

[0072] in, Indicates the kernel size as The depth of the separable convolutional layer express The first non-linear activation function operation, The purpose of feature reconstruction is to flatten the features into a one-dimensional vector. Feature scanning is used to handle long-distance dependencies between features in a flattened sequence. This indicates the normalization operation, the second one after which... The purpose of feature reconstruction is to restore features into two-dimensional vectors;

[0073] (2) For sub-features After convolution kernel size is The convolutional layer and kernel size are Depth-separable convolutional layers and Non-linear activation function operation, outputting processed sub-features The calculation formula is:

[0074] ,

[0075] in, express Nonlinear activation function operation, Indicates the kernel size as The depth of the separable convolutional layer Indicates the kernel size as Convolutional layers;

[0076] (3) Process the sub-features and The components are spliced ​​and merged, and the kernel size is [missing information]. The convolutional layer outputs features Then the features With features splicing and fusion to output features The calculation formula is:

[0077] ,

[0078] ,

[0079] in, Indicates a splicing operation;

[0080] S1.3, Features The input is fed into a downsampling layer for downsampling, and the output is satellite image feature reconstruction features. The calculation formula is:

[0081] ,

[0082] in, This indicates a downsampling operation on the features.

[0083] In a specific implementation, S2 is as follows:

[0084] The super-resolution satellite image detail enhancement module based on multi-scale convolution specifically includes three parallel convolutional layers that reconstruct features from satellite image features. Input into the super-resolution satellite image detail enhancement module;

[0085] S2.1 Reconstructing features from satellite image features The inputs are fed into three parallel convolutional layers respectively;

[0086] For the convolutional layer of the first branch: features After convolution kernel size is Depth-separable convolutional layers output features ;feature After passing through a convolution kernel with a size of Depth-separable convolutional layers output features The calculation formula is:

[0087] ,

[0088] ,

[0089] in, Indicates the kernel size as The depth of the separable convolutional layer Indicates the kernel size as The depth of the separable convolutional layer;

[0090] For the convolutional layer in the second branch: features After convolution kernel size is Convolutional layers and Nonlinear activation function, output features The calculation formula is:

[0091] ,

[0092] in, Indicates the kernel size as Convolutional layers;

[0093] For the third branch of the convolutional layer: features After convolution kernel size is The depth of the separable convolutional layer yields features and features ,feature splicing and fusion to output features ;feature After passing through a convolution kernel with a size of Depth-separable convolutional layers output features The calculation formula is:

[0094] ,

[0095] ,

[0096] in, This indicates a splicing operation. Indicates the kernel size as The depth of the separable convolutional layer Indicates the kernel size as The depth of the separable convolutional layer;

[0097] The features output by the three branches of the convolutional layer ,feature and characteristics Perform splicing and fusion to output features By stitching features together using multi-scale convolutional layers, the multi-scale representation capability of detailed features in super-resolution satellite images can be enhanced. The calculation formula is as follows:

[0098] ;

[0099] S2.2, Features After passing through a global average pooling layer and a fully connected layer, it is then combined with features. Fusion, output channel attention features Then the features After convolution kernel size is After the convolutional layer and features Fusion, output spatial attention features Then, channel attention features Spatial attention characteristics The components are spliced ​​and merged, and then processed by a convolution kernel with a size of [missing value]. The convolutional layers output super-resolution satellite image detail enhancement features. The calculation formula is:

[0100] ,

[0101] ,

[0102] ,

[0103] in, Represents broadcast multiplication of feature dimensions. This indicates a global average pooling operation. Indicates a fully connected layer. Indicates the kernel size as convolutional layers, This indicates a splicing operation.

[0104] In a specific implementation, S3 is as follows:

[0105] Enhance details in super-resolution satellite images The input is fed into the edge local perception attention module, which consists of two branches, which process the features... The two branches of the edge local perception attention module are respectively input to extract local channel features and local spatial features. Then, the two features are concatenated and fused to obtain the edge local perception attention features. ;

[0106] S3.1 Enhance details in super-resolution satellite images The features are input into the first branch of the edge local awareness attention module. The convolutional layers are sequentially processed through a global average pooling layer with a kernel size of [missing value]. convolutional layers and Activation function, output features , will feature With features Perform broadcast multiplication, then multiply the output feature with the feature... Feature concatenation is performed to ultimately output local channel features. The calculation formula is:

[0107] ,

[0108] ,

[0109] in, express Activation function operation, This indicates a global average pooling operation. This indicates a feature concatenation operation. Operators representing broadcast multiplication;

[0110] S3.2 Enhance details in super-resolution satellite images The input is fed into the second branch of the edge local awareness attention module, and the features... After convolution kernel size is The convolutional layer outputs features Then the features The input is fed into three parallel depthwise separable convolutional layers, each with a kernel size of [missing value]. The depth of the separable convolutional layer and the kernel size are The depth of the separable convolutional layer and the size of the convolutional kernel are The depthwise separable convolutional layers are used, and then the output features of the three parallel depthwise separable convolutional layers are concatenated to output features. Then the features After convolution kernel size is convolutional layers and Activation function, output features Finally, the features With features Perform broadcast multiplication, then multiply the output feature with the feature... Feature concatenation is performed to ultimately output local spatial features. The calculation formula is:

[0111] ,

[0112] ,

[0113] ,

[0114] in, Indicates the kernel size as Convolutional layers, , , These represent the kernel size as follows: The depth of the separable convolutional layer and the kernel size are The depth of the separable convolutional layer and the size of the convolutional kernel are The depth of the separable convolutional layer Operators representing broadcast multiplication;

[0115] S3.3, Local channel features With local spatial features By splicing and fusion, the edge-local awareness attention features are output. The calculation formula is:

[0116] .

[0117] In a specific implementation, step S4 is as follows:

[0118] Enhance details in super-resolution satellite images The data is input into the satellite image global capture module, which includes two branches, to process the features. The two branches of the satellite image global capture module are input separately to obtain the first branch features and the second branch features. Then, the two branch features are compared with the feature... By stitching and fusion, global features of the satellite image are obtained. ;

[0119] S4.1 Enhance details in super-resolution satellite images The data is input into the first branch of the satellite image global capture module, where features are first processed. Perform a dimension transpose operation to convert the features Rotate along the height dimension to output features Then the features The input to the convolution kernel size is The depth can be separated into convolutional layers and batch normalization layers, and the output features are then compared with the feature... Perform feature concatenation and output features. ,feature After passing through The activation function operation and the convolution kernel size are The convolutional layer outputs features ;

[0120] Features After convolution kernel size is The depth can be separated into convolutional layers and batch normalization layers, which will combine the output features with the feature set. Perform feature concatenation and output features. , will feature After passing through a convolution kernel with a size of The depth can be separable into convolutional layers, batch normalized layers, and convolutional kernels with a size of [missing value]. The convolutional layer outputs the first branch features. The calculation formula is:

[0121] ,

[0122] ,

[0123] ,

[0124] ,

[0125] ,

[0126] in, This indicates a transpose operation along the feature height dimension. The order of the dimensions is as follows , Representation of features The number of channels, Representation of features height, Representation of features The width, and the reference value of its corresponding dimension. ,feature The dimensional order is ; Indicates the batch normalization layer. Indicates the kernel size as The depth of the separable convolutional layer Indicates the kernel size as Convolutional layers;

[0127] S4.2 Enhance details in super-resolution satellite images The data is input into the second branch of the satellite image global capture module, where features are first processed. Perform a dimension transpose operation to convert the features Rotate along the width dimension to output features Then the features The input to the convolution kernel size is The depth can be separated into convolutional layers and batch normalization layers, and the obtained output is combined with the enhanced features. Perform feature concatenation and output features. Then the features go through Activation function operations and convolution kernel size are The convolutional layer outputs features ;

[0128] Enhance features After convolution kernel size is The depth can be separated into convolutional layers and batch normalization layers, which will combine the output features with the feature set. Perform feature concatenation and output features. ; Features After convolution kernel size is The depth can be separable into convolutional layers, batch normalized layers, and convolutional kernels with a size of [missing value]. The convolutional layer outputs the second branch features. The calculation formula is:

[0129] ,

[0130] ,

[0131] ,

[0132] ,

[0133] ,

[0134] in, This indicates a transpose operation along the feature width dimension. The dimensional order is ; Indicates the batch normalization layer. Indicates the kernel size as The depth of the separable convolutional layer Indicates the kernel size as Convolutional layers;

[0135] S4.3, the first branch feature Second branch features With features By stitching and fusion, global features of satellite images are output. The calculation formula is:

[0136] .

[0137] In a specific implementation, S5 is as follows:

[0138] Edge local perception attention features Global features of satellite imagery Perform splicing and fusion to output fused local and global information features. Then the features The data is input into the satellite image land category segmentation module, and after passing through a convolution kernel with a size of [missing value], it is processed. The depth can separate convolutional layers and upsampling operations, and extract features. The resolution size is restored to match the image. The same applies to images after resolution restoration. The extreme point localization operation locates the specific land category to which each pixel in an image belongs, through... Color mapping operations map the specific land category to which a pixel belongs to a different color, generating a satellite image land category segmentation image. The calculation formula is:

[0139] ,

[0140] ,

[0141] in, This indicates an upsampling operation.

[0142] Example 2

[0143] A land development category segmentation system for remote sensing images based on edge detection includes modules for executing processing instructions for each step in a land development category segmentation method for remote sensing images based on edge detection.

[0144] Example 3

[0145] To verify the effectiveness of the edge detection-based land development category segmentation method for remote sensing images proposed in this invention in the field of land category segmentation, the proposed method was compared with existing land category segmentation methods for remote sensing images under the same experimental environment and conditions. The land category segmentation results were then compared and analyzed to verify the effectiveness of the proposed method in the field of satellite remote sensing land category segmentation.

[0146] Among the existing methods for determining the high share of land categories, three mainstream segmentation methods are selected as comparative models for the method of this invention. The three comparative methods are: nnUNet model (joint neural network), nnFormer model (non-local attention-driven Transformer network), and SwinUNETR model (a network composed of Swin Transformer and three-dimensional convolutional modules).

[0147] The evaluation metrics are: F1-score, mIoU (mean intersection-over-union ratio), and OA (overall accuracy). F1-score measures the overall performance of land class segmentation by harmonic mean precision and recall. This metric is suitable for remote sensing scenarios with uneven distribution of land development categories and can comprehensively reflect the robustness of the proposed method to small sample segmentation. mIoU measures the overlap between the predicted segmented region and the ground truth labeled region, quantifying the accuracy of the proposed method in segmenting the boundaries of multiple land development targets (such as farmland, residential areas, and industrial areas). OA provides a direct assessment of the overall performance of the proposed method in land development category segmentation in complex remote sensing images.

[0148] The DeepGlobe dataset was chosen for the experiments. DeepGlobe is a publicly available high-resolution satellite remote sensing dataset for global land cover analysis and building detection. It covers various land development-related scenarios, including cities, farmland, forests, and water bodies, making it suitable for land development category segmentation and change detection tasks. The images in the DeepGlobe dataset encompass different geographical regions, seasonal features, and lighting conditions, effectively validating the segmentation performance of the proposed method in complex land cover scenarios. In the experiments, the DeepGlobe dataset was divided into training, validation, and test sets according to official standards, with a ratio of 8:1:1. This ensures the effectiveness of model training and verifies the generalization ability of the proposed method in cross-regional land development category segmentation.

[0149] The comparative experimental results of land development category segmentation based on miniature remote sensing images are shown in Table 1. The experimental results show that the proposed method significantly outperforms the mainstream comparative model in the land development category segmentation task. Its F1 score reaches 89.74%, the mean intersection-over-union ratio (mIoU) is 85.15%, and the overall accuracy (OA) is 89.06%. These three indicators are improved by 3.86%, 2.95%, and 5.38% respectively compared with the best comparative model SwinUNETR.

[0150] Table 1 shows the comparison results of validating the proposed method on the dataset.

[0151]

[0152] Figure 2 To more intuitively compare the segmentation effect of the method proposed in this invention with other methods, Figure 2 The purple areas represent buildings, the black areas represent forests, and the light blue areas represent low vegetation areas. Figure 2 As can be seen, the method proposed in this invention is closer to the actual land remote sensing image classification region in the actual segmentation effect;

[0153] Therefore, the above experimental results demonstrate the effectiveness and advancement of the proposed method in the field of land development category segmentation in satellite remote sensing images.

[0154] Although the specific embodiments of the invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the invention. Based on the technical solutions of the invention, various modifications or variations that can be made by those skilled in the art without creative effort are still within the scope of protection of the invention.

Claims

1. A method for land development category segmentation in remote sensing images based on edge detection, characterized in that, A land development category segmentation model based on satellite remote sensing images is constructed. This model is input into satellite remote sensing images of land development to obtain the land development category segmentation results. The steps include: S1. Construct a satellite image feature reconstruction module: This module includes depthwise separable convolution operations. The activation function, satellite image feature space reconstruction module, and downsampling layer input satellite remote sensing land images captured by satellite remote sensing into the satellite image feature reconstruction module and output satellite image feature reconstruction features. S1 is as follows: S1.1 Satellite remote sensing land images The kernel size input to the satellite image feature reconstruction module is Depth separable convolutional layers and Activation function, output features ; S1.2, Features The satellite image feature space reconstruction module, which inputs the data into the satellite image feature reconstruction module, first modifies the features through a transpose operation. The spatial structure, after transpose operation, outputs transformed features. Transformation characteristics After passing through a convolution kernel with a size of The convolutional layer is then processed by feature segmentation. Operations will transform features The number of channels segmented along the channel direction is a transformation feature. sub-features and ; (1) For sub-features This feature is processed by a convolution kernel with a size of Depth-separable convolution operations Nonlinear activation function operations Feature reconstruction operation Feature scanning operation Normalization operation, and reuse Reconstruction operation, outputting processed sub-features The first time I passed by The feature reconstruction operation flattens the features into a one-dimensional vector, and the second pass... Feature reconstruction operations restore features to two-dimensional vectors; (2) For sub-features After convolution kernel size is The convolutional layer and kernel size are Depth-separable convolutional layers and Non-linear activation function operation, outputting processed sub-features ; (3) Process the sub-features and The components are spliced ​​and merged, and the kernel size is [missing information]. The convolutional layer outputs features Then the features With features splicing and fusion to output features ; S1.3, Features The input is fed into a downsampling layer for downsampling, and the output is satellite image feature reconstruction features. ; S2. Construct a super-resolution satellite image detail enhancement module: Construct a super-resolution satellite image detail enhancement module based on multi-scale convolution. Input the satellite image feature reconstruction features into the super-resolution satellite image detail enhancement module to perform super-resolution detail enhancement, and output the super-resolution satellite image detail enhancement features. S2 is as follows: The super-resolution satellite image detail enhancement module based on multi-scale convolution specifically includes three parallel convolutional layers that reconstruct features from satellite image features. Input into the super-resolution satellite image detail enhancement module; S2.1 Reconstructing features from satellite image features The inputs are fed into three parallel convolutional layers respectively; For the convolutional layer of the first branch: features After convolution kernel size is Depth-separable convolutional layers output features ;feature After passing through a convolution kernel with a size of Depth-separable convolutional layers output features ; For the convolutional layer in the second branch: features After convolution kernel size is Convolutional layers and Nonlinear activation function, output features ; For the third branch of the convolutional layer: features After convolution kernel size is The depth of the separable convolutional layer yields features and features ,feature splicing and fusion to output features ;feature After passing through a convolution kernel with a size of Depth-separable convolutional layers output features ; The features output by the three branches of the convolutional layer ,feature and characteristics Perform splicing and fusion to output features By stitching features together using multi-scale convolutional layers, the multi-scale representation capability of detailed features in super-resolution satellite images can be enhanced. S2.2, Features After passing through a global average pooling layer and a fully connected layer, it is then combined with features. Fusion, output channel attention features Then the features After convolution kernel size is After the convolutional layer and features Fusion, output spatial attention features Then, channel attention features Spatial attention characteristics The components are spliced ​​and merged, and then processed by a convolution kernel with a size of [missing value]. The convolutional layers output super-resolution satellite image detail enhancement features. ; S3. Construct an edge local perception attention module: This module includes an edge local perception attention module, which inputs the super-resolution satellite image detail enhancement features into the edge local perception attention module for local feature capture and outputs edge local perception attention features. S3 is as follows: Enhance details in super-resolution satellite images The input is fed into the edge local perception attention module, which consists of two branches, which process the features... The two branches of the edge local perception attention module are respectively input to extract local channel features and local spatial features. Then, the two features are concatenated and fused to obtain the edge local perception attention features. ; S3.1 Enhance details in super-resolution satellite images The features are input into the first branch of the edge local awareness attention module. The convolutional layers sequentially pass through a global average pooling layer with a kernel size of [missing value]. convolutional layers and Activation function, output features , will feature With features Perform broadcast multiplication, then multiply the output features by the feature... Feature concatenation is performed to ultimately output local channel features. ; The calculation formula is: ， ， in, express Activation function operation, This indicates a global average pooling operation. This indicates a feature concatenation operation. Operators representing broadcast multiplication; S3.2 Enhance details in super-resolution satellite images The input is fed into the second branch of the edge local awareness attention module, and the features... After convolution kernel size is The convolutional layer outputs features Then the features The input is fed into three parallel depthwise separable convolutional layers, each with a kernel size of [missing value]. The depth of the separable convolutional layer and the kernel size are The depth of the separable convolutional layer and the size of the convolutional kernel are The depthwise separable convolutional layers are used, and then the output features of the three parallel depthwise separable convolutional layers are concatenated to output features. Then the features After convolution kernel size is convolutional layers and Activation function, output features Finally, the features With features Perform broadcast multiplication, then multiply the output feature with the feature... Feature concatenation is performed to ultimately output local spatial features. ; S3.3, Local channel features With local spatial features By splicing and fusion, the edge-local awareness attention features are output. ; S4. Construct a global satellite image capture module: This module includes a global satellite image capture module, which inputs edge local perception attention features into the global satellite image capture module to enhance the global correlation of features and outputs global satellite image features; S4 is as follows: Enhance details in super-resolution satellite images The data is input into the satellite image global capture module, which includes two branches, to process the features. The two branches of the satellite image global capture module are input separately to obtain the first branch features and the second branch features. Then, the two branch features are compared with the feature... By stitching and fusion, global features of the satellite image are obtained. ; S4.1 Enhance details in super-resolution satellite images The data is input into the first branch of the satellite image global capture module, where features are first processed. Perform a dimension transpose operation to convert the features Rotate along the height dimension to output features Then the features The input to the convolution kernel size is The depth can be separated into convolutional layers and batch normalization layers, and the output features are then compared with the feature... Perform feature concatenation and output features. ,feature After passing through The activation function operation and the convolution kernel size are The convolutional layer outputs features ; Features After convolution kernel size is The depth can be separated into convolutional layers and batch normalization layers, which will combine the output features with the feature set. Perform feature concatenation and output features. , will feature After passing through a convolution kernel with a size of The depth can be separable into convolutional layers, batch normalized layers, and convolutional kernels with a size of [missing information]. The convolutional layer outputs the first branch features. ; The calculation formula is: ， ， ， ， ， in, This indicates a transpose operation along the feature height dimension. The order of the dimensions is as follows , Representation of features The number of channels, Representation of features height, Representation of features The width, and the reference value of its corresponding dimension. ,feature The dimensional order is ; Indicates the batch normalization layer. Indicates the kernel size as The depth of the separable convolutional layer Indicates the kernel size as Convolutional layers; S4.2 Enhance details in super-resolution satellite images The data is input into the second branch of the satellite image global capture module, where features are first processed. Perform a dimension transpose operation to convert the features Rotate along the width dimension to output features Then the features The input to the convolution kernel size is The depth can be separated into convolutional layers and batch normalization layers, and the obtained output is combined with the enhanced features. Perform feature concatenation and output features. Then the features go through Activation function operations and convolution kernel size are The convolutional layer outputs features ; Enhance features After convolution kernel size is The depth can be separated into convolutional layers and batch normalization layers, which will combine the output features with the feature set. Perform feature concatenation and output features. ; Features After convolution kernel size is The depth can be separable into convolutional layers, batch normalized layers, and convolutional kernels with a size of [missing value]. The convolutional layer outputs the second branch features. ; The calculation formula is: ， ， ， ， ， in, This indicates a transpose operation along the feature width dimension. The dimensional order is ; Indicates the batch normalization layer. Indicates the kernel size as The depth of the separable convolutional layer Indicates the kernel size as Convolutional layers; S4.3, the first branch feature Second branch features With features By stitching and fusion, global features of satellite images are output. ; S5. Construct a satellite image land category segmentation module: Combine local edge awareness attention features with global satellite image features through feature concatenation to output fused local and global information features. Then, input these features into the satellite image land category segmentation module to output a satellite image land category segmented image. .

2. The method for land development category segmentation in remote sensing images based on edge detection according to claim 1, characterized in that, S5 is detailed below: Edge local perception attention features Global features of satellite imagery Perform splicing and fusion to output fused local and global information features. Then the features The data is input into the satellite image land category segmentation module, and after passing through a convolution kernel with a size of [missing value], it is processed. The depth can separate convolutional layers and upsampling operations, and extract features. The resolution size is restored to match the image. The same applies to images after resolution restoration. The extreme point localization operation locates the specific land category to which each pixel in an image belongs, through... Color mapping operations map the specific land category to which a pixel belongs to a different color, generating a satellite image land category segmentation image. .

3. A land development category segmentation system for remote sensing images based on edge detection, characterized in that: It includes a module for executing the processing instructions for each step in the remote sensing image land development category segmentation method based on edge detection as described in any one of claims 1-2.

Images