Erosion gully segmentation method based on multi-scale edge perception

Abstract

The invention discloses an erosion gully segmentation method based on multi-scale edge perception, and relates to the technical field of remote sensing image target extraction. The invention aims to solve the problem that the boundary extraction of the erosion gully by a common neural network algorithm is not clear due to the complicated geographical environment and characteristics of the erosion gully. According to the erosion gully segmentation method based on multi-scale edge perception, a boundary detection module and a multi-scale decoder are adopted, the edge detection module is used for extracting edge features of an erosion gully, the capability of extracting boundary details of the erosion gully is enhanced, and the segmentation precision of the whole erosion gully is improved; the multi-scale decoder decodes the features of the encoder and the boundary detection module, context feature information extracted by the encoder is fully utilized, the loss of detail feature information is inhibited, and the segmentation capability for fuzzy gullies and the overall segmentation precision are improved.

Description

Technical Field

[0001] The invention belongs to the technical field of remote sensing image target extraction. Background Art

[0002] Gullies are formed when the surface soil is eroded by wind and water, which can lead to soil loss, land degradation, ecosystem destruction and other disasters. In recent years, due to the intensification of environmental pollution and the increasing number of extreme weather conditions, the number and formation rate of gullies have increased accordingly. Therefore, effective detection and prevention measures for gullies can effectively protect land resources and maintain the sustainable development of the ecological environment. Therefore, regular monitoring of gullies is crucial to understand their evolution and formulate effective restoration strategies.

[0003] For the extraction of erosion gullies, the traditional method is to determine the erosion gullies through manual field surveys or manual interpretation of satellite images. Although this method is accurate, it is time-consuming, labor-intensive and costly. Later, remote sensing images were used to detect erosion gullies using manual visual interpretation or traditional machine learning methods. This method has improved efficiency, but the accuracy is very low.

[0004] In recent years, with the emergence of deep learning neural network methods, supervised semantic segmentation methods based on convolutional neural networks have attracted extensive research. This method uses high-resolution satellite images as input, uses deep learning technology, and passes through semantic segmentation neural networks to obtain binary segmentation results (background and gullies), such as Figure 1 As shown. This method can automatically learn more powerful hierarchical feature representations from raw data, providing new possibilities for automatic monitoring of erosion gullies. For example, Xu Zhaopeng et al. proposed a method based on a fully convolutional network for erosion gully monitoring in the Loess Plateau of China. By combining remote sensing images with high-resolution digital terrain models and using FCN networks for semantic segmentation, automatic monitoring of erosion gullies can be achieved. Zhou Jian et al. proposed an automatic extraction method based on a back-propagation neural network. Using a digital elevation model and based on feature analysis of loess erosion gullies, accurate gully extraction results were obtained. Zhao Chunhui et al. proposed a semi-supervised semantic segmentation method based on a boundary-guided pseudo-label generation strategy and an adaptive loss function method, which can make full use of labeled historical data and the latest unlabeled monitoring data to obtain the latest erosion gully monitoring results.

[0005] Although some achievements and progress have been made in gully monitoring, most of the current methods simply apply some segmentation networks in the field of computer vision (such as FCN and U-Net, etc.) to the task of gully extraction. However, compared with traditional visual images, due to the complex geographical environment and characteristics of gullies, the representation of gullies on remote sensing images is diverse, and the surrounding environmental characteristics are also different. This leads to the general neural network algorithm for gully boundary extraction is unclear, especially for small-scale gullies. The extraction effect is very poor and it is difficult to achieve the expected extraction effect. Summary of the invention

[0006] The present invention aims to solve the problem that the general neural network algorithm is unclear about the boundary extraction of erosion gullies due to the complex geographical environment and characteristics of erosion gullies. A erosion gully segmentation method based on multi-scale edge perception is now provided.

[0007] A multi-scale edge-aware erosion gully segmentation method is proposed, where the erosion gully image is input into a semantic segmentation network to obtain the erosion gully and background segmentation images.

[0008] The semantic segmentation network includes: an encoder, a boundary detection module and a multi-scale decoder;

[0009] The erosion gully image is used as the input of the boundary detection module and the encoder respectively, the boundary map of the erosion gully image is obtained by the boundary detection module, and the feature map of the erosion gully image is obtained by the encoder;

[0010] The multi-scale decoder is used to perform channel integration on the boundary map and the feature map, and the overall features of the erosion gully image are obtained as the segmentation result, thereby completing the segmentation of the erosion gully image.

[0011] Furthermore, the above-mentioned boundary detection module includes: two 3×3 convolutional layers, a regularization layer and a 2×2 maximum pooling layer, and the features output by the previous layer are used as the input of the next layer.

[0012] Furthermore, the output features of the four layers in the above boundary detection module are expressed as:

[0013]

[0014] Wherein, R represents a real number, H and W represent the height and width of the erosion groove image respectively.

[0015] Furthermore, the above-mentioned boundary detection module also includes seven boundary enhancement modules, four of which constitute the first-level boundary enhancement modules, and the remaining three constitute the second-level boundary enhancement modules;

[0016] The output features of the four layers of the boundary detection module are respectively input into four first-level boundary enhancement modules, the output of the first-level boundary enhancement module of the first layer and the output of the second-level boundary enhancement module of the second layer are used together as the input of the second-level boundary enhancement module of the second layer, the output of the second-level boundary enhancement module of the second layer and the output of the third-level boundary enhancement module of the first layer are used together as the input of the third-level boundary enhancement module, and the output of the third-level boundary enhancement module and the output of the fourth-level boundary enhancement module of the first layer are used together as the input of the fourth-level boundary enhancement module.

[0017] Furthermore, the above-mentioned boundary enhancement module includes: a Sobel convolution layer, a normalization layer, an activation function layer, a 3*3 convolution layer and a 1*1 convolution layer;

[0018] The input of the boundary enhancement module passes through the Sobel convolution layer, the normalization layer, the activation function layer and the 3*3 convolution layer in sequence. The output of the 3*3 convolution layer and the input of the boundary enhancement module pass through the 1*1 convolution layer together to output the boundary map of the erosion gully image.

[0019] Furthermore, the multi-scale decoder includes: two 1×1 convolutional layers, one 3×3 convolutional layer, two 3×3 dilated convolutional layers and a spatial attention module, wherein the expansion rates of the two 3×3 dilated convolutional layers are 2 and 3 respectively;

[0020] The boundary map and the feature map are used together as input features of the multi-scale decoder. The input features are respectively passed through a 1×1 convolution layer, a 3×3 convolution layer and two 3×3 hole convolution layers to obtain four output results. Another 1×1 convolution layer integrates the four output results at the channel level, and then the features are further enhanced through the spatial attention module to obtain the overall features of the erosion gully image.

[0021] Furthermore, the output of a 1×1 convolutional layer is Y1=Conv1*1(X);

[0022] The output of the 3×3 convolutional layer is Y2=Conv3*3(X);

[0023] The output of the 3×3 dilated convolutional layer with a dilation rate of 2 is Y3=DilatedConv3*3(X,2);

[0024] The output of the 3×3 dilated convolutional layer with a dilation rate of 3 is Y4=DilatedConv3*3(X,3);

[0025] X represents the input feature, Conv1*1() represents a 1×1 convolution operation, Conv3*3(X) represents a 3×3 convolution operation, and DilatedConv3*3() represents a 3×3 dilated convolution operation.

[0026] Furthermore, the above encoder is an encoder of the Segformer algorithm.

[0027] Furthermore, the method for obtaining the above erosion gully image is:

[0028] Collecting remote sensing images of the target area through satellites, and cropping and scaling the remote sensing images in sequence so that the remote sensing images can adapt to the processing conditions of the semantic segmentation network;

[0029] The adjusted remote sensing image is then standardized to obtain the erosion gully image.

[0030] Furthermore, the above standardized expression is as follows:

[0031]

[0032] Among them, x′ represents the normalized pixel value, x i is the pixel value of the i-th pixel in the adjusted remote sensing image, i=1,2,...,N, N is the total number of pixels in the adjusted remote sensing image, The average value of all pixels in the adjusted remote sensing image.

[0033] The present invention proposes a gully segmentation method based on multi-scale edge perception, which adopts a boundary detection module (Boundary Detection module, BDM) and a multi-scale decoder (Multiscale Perception Decoder, MPD), wherein the edge detection module extracts the edge features of the gully, enhances the ability to extract the details of the gully boundary, and improves the overall segmentation accuracy of the gully; the multi-scale decoder decodes the features of the encoder and the boundary detection module, makes full use of the context feature information extracted by the encoder, suppresses the loss of detail feature information, and improves the segmentation ability and overall segmentation accuracy of fuzzy gullies. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A schematic diagram of a method for segmenting an erosion gully image using deep learning technology as described in the background technology;

[0035] Figure 2 This is a schematic diagram of the architecture principle of the Segformer network;

[0036] Figure 3 It is a structural diagram of the boundary detection module;

[0037] Figure 4 It is a structural diagram of the boundary enhancement module;

[0038] Figure 5 It is a schematic diagram of the structure of a multi-scale decoder;

[0039] Figure 6 This is a data picture of the erosion gully in the second specific implementation mode;

[0040] Figure 7 This is the visualization result obtained by using different gully segmentation algorithms;

[0041] Figure 8 Comparison chart of visualization results after improving the algorithm. DETAILED DESCRIPTION

[0042] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention. It should be noted that the embodiments of the present invention and the features in the embodiments can be combined with each other without conflict. Specific implementation method one:

[0044] Gullies are narrow, deep and steep gullies formed during soil erosion. Their existence brings a series of hazards to soil, water resources and ecological environment. Formulating efficient gully monitoring measures will help prevent soil and water loss and achieve sustainable development of land resources. The semantic segmentation method based on deep learning can automatically extract gullies from remote sensing images of the area to be monitored and is generally considered to be the most advanced method. However, how to extract reliable gully results from complex natural environments in the wild is a challenging task. To this end, this embodiment proposes a semantic segmentation algorithm based on multi-scale edge perception. This method obtains more refined gully segmentation results by effectively acquiring detail information and edge information between different terrains.

[0045] Reference Figures 2 to 5 Specifically describing this embodiment, the erosion gully segmentation method based on multi-scale edge perception described in this embodiment collects remote sensing images of the target area through satellites, and the remote sensing images are cropped and scaled in sequence so that the remote sensing images can adapt to the processing conditions of the semantic segmentation network. Then, the adjusted remote sensing images are standardized to obtain erosion gully images. The standardized expression is as follows:

[0046]

[0047] Among them, x′ represents the normalized pixel value, x i is the pixel value of the i-th pixel in the adjusted remote sensing image, i=1,2,...,N, N is the total number of pixels in the adjusted remote sensing image, The average value of all pixels in the adjusted remote sensing image.

[0048] The erosion gully image is input into a semantic segmentation network to obtain an erosion gully and background segmentation image. The semantic segmentation network includes: an encoder, a boundary detection module and a multi-scale decoder.

[0049] The erosion gully image is used as the input of the boundary detection module and the encoder respectively. The boundary map of the erosion gully image is obtained by the boundary detection module, and the feature map of the erosion gully image is obtained by the encoder.

[0050] The multi-scale decoder is used to perform channel integration on the boundary map and the feature map to obtain the overall features of the erosion gully image.

[0051] The overall features of the erosion gully image are up-sampled to obtain an erosion gully and background segmentation image having the same size as the erosion gully image.

[0052] 1. Semantic Segmentation Network

[0053] In recent years, many methods based on convolutional neural networks (CNN) and Transformer structures have emerged in the field of semantic segmentation. Convolutional neural networks (CNN) have achieved remarkable success in semantic segmentation tasks. CNN can effectively capture local and global features through convolution operations and has hierarchical feature representation capabilities. In addition, methods based on the Transformer structure have also attracted widespread attention in the field of semantic segmentation. Compared with the CNN algorithm, the Transformer structure can better capture global information and has better long-range dependency modeling capabilities, and has more outstanding performance in semantic segmentation tasks.

[0054] The existing technology applies the core ideas and architecture of the Transformer model to pixel-level image segmentation tasks. Its main structural innovations are as follows:

[0055] Self-Attention: The self-attention mechanism enables the model to weight the importance of each position in the input sequence when generating feature representations. In the semantic segmentation task, the self-attention mechanism is applied to the spatial dimension of the image, allowing the model to model the correlation between different pixels.

[0056] Feed-Forward Neural Network: Feed-forward neural network performs independent nonlinear transformation on the features of each position. It is used with self-attention mechanism to build a deep network by stacking multiple layers.

[0057] Multi-scale processing: Transformer-based semantic segmentation algorithms usually introduce multi-scale processing mechanisms. By processing features at different levels, the algorithm can capture contextual information at different scales. Common methods include pyramid pooling and multi-scale attention mechanisms.

[0058] This implementation adopts the Segformer algorithm, a semantic segmentation algorithm based on the Transformer structure. Compared with other semantic segmentation algorithms with the Transformer structure, the SegFormer algorithm has advantages in multi-scale processing capabilities, dual-stream decoders, cross-stage connection characteristics, and efficient scalability in the erosion gully segmentation task. It can better cope with the challenges of the erosion gully segmentation task and achieve more accurate segmentation results.

[0059] The encoder part of the Segformer algorithm is composed of multiple Transformer blocks and input embedding and encoding modules. Each Transformer block contains a self-attention mechanism for calculating the attention weight between each position and other positions and a feedforward neural network for nonlinear transformation and extraction of features. The input embedding and position encoding module is used to convert the pixel values ​​in the input image into a feature vector representation and add position information to each position in the input sequence so that the model can distinguish features at different positions. Therefore, the encoder in this embodiment is the encoder of the Segformer algorithm.

[0060] 2. Boundary detection module

[0061] In gully landforms, the surrounding environment is complex and diverse. The network usually performs poorly at the edge where the gully meets the ground object, which leads to poor overall segmentation accuracy and inaccurate prediction. To address this problem, this implementation has designed a boundary detection module (Boundary Detection Module, BDM) to improve the detection capability of gully boundaries by extracting image boundary features. The structure of the edge detection module is as follows: Figure 3 As shown:

[0062] To enhance the extraction of image boundary features, after the erosion gully image is input into the boundary detection block, it passes through four convolution blocks to extract features of different dimensions. The four convolution blocks are two 3×3 convolution layers, a regularization layer and a 2×2 maximum pooling layer. The features output by the previous layer are used as the input of the next layer.

[0063] The output features of the four layers are expressed as:

[0064]

[0065] Wherein, R represents a real number, H and W represent the height and width of the erosion groove image respectively.

[0066] The boundary detection module also includes seven boundary enhancement modules, four of which constitute first-level boundary enhancement modules, and the remaining three constitute second-level boundary enhancement modules; the output features of the four layers of the boundary detection module are respectively input into the four first-level boundary enhancement modules, the output of the first-level boundary enhancement module of the first layer and the output of the first-level boundary enhancement module of the second layer are used together as the input of the second-level boundary enhancement module of the second layer, the output of the second-level boundary enhancement module of the second layer and the output of the first-level boundary enhancement module of the third layer are used together as the input of the second-level boundary enhancement module of the third layer, and the output of the third-level boundary enhancement module and the output of the fourth-level boundary enhancement module of the first layer are used together as the input of the fourth-level boundary enhancement module.

[0067] In this embodiment, the features of four sizes obtained by the four convolutional layers are respectively input into the corresponding boundary enhancement modules, and the edge space is extracted by the BEM module. The specific structure of the BEM module is as follows: Figure 4 shown.

[0068] The boundary enhancement module includes: Sobel convolution layer, normalization layer, activation function layer, 3*3 convolution layer and 1*1 convolution layer; a residual structure system is used to enhance the extraction of boundary features, the input is extracted through the Sobel (Sobel operator) convolution layer and the 3*3 convolution layer, and the residual is reduced to a single channel through the 1*1 convolution layer, and the boundary feature map is output. Specifically, the input of the boundary enhancement module passes through the Sobel convolution layer, normalization layer, activation function layer and 3*3 convolution layer in sequence, and the output of the 3*3 convolution layer and the input of the boundary enhancement module pass through the 1*1 convolution layer to output the boundary map of the erosion gully image. Richer edge features are obtained by progressive superposition, and finally the features are interpolated to the original image size to reconstruct the boundary detection results. Through the boundary detection module, the boundary information of the erosion gully in the image can be effectively extracted, which helps the model to accurately segment the erosion gully area.

[0069] 3. Multi-scale decoder

[0070] Although the boundary detection module performs well in erosion gully boundary segmentation, it still performs poorly in some small erosion gullies, often missing or misdetecting. For these small gullies, context information is very important for small-scale erosion gully segmentation. To address this situation, this embodiment designs a multi-scale perception decoder, whose overall structure is as follows: Figure 5As shown, the multi-scale decoder includes: two 1×1 convolutional layers, one 3×3 convolutional layer, two 3×3 hole convolutional layers and a spatial attention module, and the expansion rates of the two 3×3 hole convolutional layers are 2 and 3 respectively.

[0071] The overall decoder adopts pyramid convolution. The features extracted by the encoder are passed through dilated convolution and pooling layers with different dilation rates. The dilated convolution increases the decoder's receptive field and extracts semantic information of different scales. The spatial pyramid pooling operation is applied to divide the feature map into sub-regions of different scales, and each sub-region is pooled. This operation helps to extract richer semantic information while maintaining the spatial structure of the feature map. Finally, 1*1 convolution is used to integrate the multi-layer features at the channel level, and then a spatial attention module is used to further enhance the features and output the results.

[0072] Specifically, the boundary map and the feature map are used as input features of the multi-scale decoder, and the input features are respectively passed through a 1×1 convolution layer, a 3×3 convolution layer and two 3×3 hole convolution layers to obtain four output results. The number of output channels of the four layers is C1, C2, C3, and C4 respectively.

[0073] The output of a 1×1 convolutional layer is Y1=Conv1*1(X);

[0074] The output of the 3×3 convolutional layer is Y2=Conv3*3(X);

[0075] The output of the 3×3 dilated convolutional layer with a dilation rate of 2 is Y3=DilatedConv3*3(X,2);

[0076] The output of the 3×3 dilated convolutional layer with a dilation rate of 3 is Y4=DilatedConv3*3(X,3);

[0077] X represents the input feature, Conv1*1() represents a 1×1 convolution operation, Conv3*3(X) represents a 3×3 convolution operation, and DilatedConv3*3() represents a 3×3 dilated convolution operation.

[0078] Another 1×1 convolutional layer integrates the four output results at the channel level, and then further enhances the features through the spatial attention module to obtain the overall features of the erosion gully image.

[0079] The number of channels C that are finally input into the spatial attention module Out for:

[0080] C Out =C1+C2+C3+C4.

[0081] Through the multi-scale perception decoder, the contextual information and detail information in the features can be effectively perceived and extracted, and the model's ability to extract small gullies can be enhanced.

[0082] In summary, this implementation adopts the architecture of the Segformer network as a whole. The encoder part adds a boundary detection module (BDM) in parallel with the original encoder, aiming to extract edge information in the image and generate a boundary map, so that the features extracted by the model can obtain more boundary information of erosion gullies. The decoder part adds a multiscale perception decoder (MPD) to perform multi-scale extraction on the feature information extracted by the encoder, maintain context information and detailed edge information, and capture more details of fuzzy gullies, thereby assisting the model to perform more detailed erosion gully segmentation tasks. Specific implementation method 2:

[0084] The gully data used in this implementation is the data taken by the GF-2 satellite in Huachuan County, Heilongjiang Province. The data band is 3, and the processed data images are cut into more than 1,000 384*384 slices. The data images containing gullies are manually annotated, and the annotated categories are: gullies and background, and different colors are used to distinguish them. More than a thousand images are randomly divided into training sets and test sets with a ratio of 7:3. The data images are as follows: Figure 6 shown.

[0085] There are two main evaluation criteria for gully segmentation. Pixel accuracy Acc: Pixel accuracy refers to the ratio of the number of pixels predicted correctly by the model to the total number of pixels. It is a simple and intuitive evaluation indicator, but it may not reflect the situation of class imbalance. The second is Mean Intersection over Union (mIoU): mIoU is a commonly used evaluation criterion that calculates the intersection over union (IoU) of each category and then takes the average of all categories. IoU is the ratio of the intersection area to the union area between the predicted result and the true label, which can measure the precision and accuracy of the predicted result.

[0086] The specific calculation formula is as follows:

[0087]

[0088] Among them, TP represents the number of pixels correctly predicted as erosion gullies, TN represents the number of pixels correctly predicted as non-erosion gullies, FP represents the number of pixels incorrectly predicted as non-erosion gullies, and FN represents the number of pixels incorrectly predicted as non-erosion gullies.

[0089] On the Huachuan erosion gully dataset, this implementation uses a variety of classic semantic segmentation algorithms to train and test the dataset, and obtains the following results as shown in Table 1:

[0090] Table 1: Results of classic semantic segmentation algorithms

[0091]

[0092] As can be seen from the table, due to the complex geomorphic environment of the erosion gullies, the accuracy of most traditional CNN semantic segmentation methods is relatively low, with IoU around 40, and the best HrNet reached 43.14. The reason for the low accuracy of CNN segmentation may be that the traditional CNN semantic segmentation algorithm is difficult to accurately capture the different scales and shapes of the erosion gullies, and lacks information when processing small-scale details. For the segmentation algorithm based on the Tansformer structure, the accuracy is generally higher than the semantic segmentation algorithm of the CNN structure. Because it has a certain control over multi-scale details, it can capture the small-scale information of the erosion gullies to a certain extent. Among the segmentation algorithms based on the Tansformer structure, the Segformer algorithm has the highest accuracy, with an IoU of 50.34.

[0093] Due to the complex geomorphic environment of erosion gullies, the classic traditional semantic segmentation algorithm has poor segmentation effect on gullies. The Segformer algorithm based on the Transformer structure performs better, but its visualization results still have many problems. The visualization results are as follows: Figure 7 As shown. Red pixels represent false detections, and blue pixels represent missed detections. From the visualization results, we can see that for obvious gullies such as (a)(b)(c), some CNN methods still have large missed detection problems, and the segmentation results of the gullies are not accurate enough. The Transformer algorithm performs better than the CNN algorithm, but there are still some missed detections. From the method of this embodiment, we can see that the segmentation of these simple gullies is more accurate, with fewer missed detections and false detections. For complex and large gullies such as (d)(e), we can see that the results of the CNN method have large missed detections and false detections, resulting in broken segmentation results, and the Transformer method does not perform well. However, from the method of this embodiment, we can see that the segmentation continuity and accuracy of the erosion gullies have reached a high level, which is particularly prominent in Figure (e).

[0094] In order to improve the accuracy of erosion gully segmentation, improve the boundary segmentation effect and reduce the missed detection rate, this implementation uses Segformer as the base algorithm, connects the boundary segmentation module in parallel with the original encoder, and enriches the feature information extracted by the encoder. The decoder part uses a multi-scale perception decoder to enrich the model's extraction of multi-scale information. The ablation experiment was performed on the original data set, and the results are shown in Table 2:

[0095] Table 2: Improved experimental accuracy

[0096]

[0097] As shown in Table 2, when the boundary detection module and multi-scale decoder are added respectively, the accuracy of the model is improved by 3.71% and 2.5% respectively. When both are added at the same time, the accuracy is improved most significantly, reaching 6.76%. This shows that the multi-scale decoder can more accurately extract the features extracted by the boundary detection module, assist the model in segmentation, and obtain more accurate segmentation results. After adding the boundary detection module and the multi-scale perception decoder, (1) the model's detection ability for small erosion gullies is significantly enhanced. The missed detection of small erosion gullies in the original results is improved in the improved algorithm, which can be seen in the results (a) and (b), effectively improving the segmentation accuracy of erosion gullies. (2) In the edge areas of some large erosion gullies, due to the addition of boundary detection blocks, the segmentation effect is significantly improved, and the edge segmentation is more detailed and accurate. This improvement can be seen in the results (a) and (c). Overall, compared with the results of traditional semantic segmentation algorithms, Base+EDM+MPD can effectively improve the segmentation accuracy of gully boundaries, improve the missed detection of small gullies, and effectively improve segmentation accuracy, providing reliable technical support for obtaining gully disaster information and formulating effective prevention and control plans.

[0098] In this embodiment, the data image is input into the network, the feature information of the image is extracted by the encoder in cooperation with the boundary detection module, and the feature information is decoded more deeply in combination with the multi-scale decoder to obtain a more accurate gully detection result. Among them, the boundary detection module EDM proposed in the semantic segmentation method can improve the model's detection effect on the gully boundary. After adding it, the accuracy is improved by 3.71% compared with the base network. The proposed multi-scale perception decoder MPD can improve the model's detection ability for small gullies. After adding it, the model accuracy is improved by 2.5%. After both are added to the network at the same time, the improvement reaches 6.76%. Compared with other traditional algorithms, the method proposed in this article can effectively improve the difficult problems in gully detection, improve the segmentation accuracy of gully detection tasks, and provide effective information for the prevention and control of gullies.

[0099] Although the present invention is described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the present invention. It should therefore be understood that many modifications may be made to the exemplary embodiments and that other arrangements may be devised without departing from the spirit and scope of the present invention as defined by the appended claims. It should be understood that the various dependent claims and features described herein may be combined in a manner different from that described in the original claims. It should also be understood that the features described in conjunction with a single embodiment may be used in other described embodiments.

Claims

1. A gully segmentation method based on multi-scale edge perception, which inputs the gully image into the semantic segmentation network to obtain the gully and background segmentation image; It is characterized in that The semantic segmentation network includes: an encoder, a boundary detection module and a multi-scale decoder; The erosion gully image is used as the input of the boundary detection module and the encoder respectively, the boundary map of the erosion gully image is obtained by the boundary detection module, and the feature map of the erosion gully image is obtained by the encoder; The multi-scale decoder is used to perform channel integration on the boundary map and the feature map, and the overall features of the erosion gully image are obtained as the segmentation result, thereby completing the segmentation of the erosion gully image.

2. The multi-scale edge-aware erosion gully segmentation method according to claim 1, characterized in that: The boundary detection module includes: two 3×3 convolutional layers, a regularization layer and a 2×2 maximum pooling layer, and the features output by the previous layer are used as the input of the next layer.

3. The method for erosion gully segmentation based on multi-scale edge perception according to claim 2, characterized in that: The output features of the four layers in the boundary detection module are respectively expressed as: Wherein, R represents a real number, H and W represent the height and width of the erosion groove image respectively.

4. The method for gully segmentation based on multi-scale edge perception according to claim 2 or 3, characterized in that: The boundary detection module also includes seven boundary enhancement modules, four of which constitute primary boundary enhancement modules and the remaining three constitute secondary boundary enhancement modules; The output features of the four layers of the boundary detection module are respectively input into four first-level boundary enhancement modules, the output of the first-level boundary enhancement module of the first layer and the output of the second-level boundary enhancement module of the second layer are used together as the input of the second-level boundary enhancement module of the second layer, the output of the second-level boundary enhancement module of the second layer and the output of the third-level boundary enhancement module of the first layer are used together as the input of the third-level boundary enhancement module, and the output of the third-level boundary enhancement module and the output of the fourth-level boundary enhancement module of the first layer are used together as the input of the fourth-level boundary enhancement module.

5. The multi-scale edge-aware erosion gully segmentation method according to claim 4, characterized in that: The boundary enhancement module includes: a Sobel convolution layer, a normalization layer, an activation function layer, a 3*3 convolution layer and a 1*1 convolution layer; The input of the boundary enhancement module passes through the Sobel convolution layer, the normalization layer, the activation function layer and the 3*3 convolution layer in sequence. The output of the 3*3 convolution layer and the input of the boundary enhancement module pass through the 1*1 convolution layer together to output the boundary map of the erosion gully image.

6. The multi-scale edge-aware erosion gully segmentation method according to claim 1, 2, 3 or 5, characterized in that: The multi-scale decoder includes: two 1×1 convolutional layers, one 3×3 convolutional layer, two 3×3 dilated convolutional layers and a spatial attention module, wherein the expansion rates of the two 3×3 dilated convolutional layers are 2 and 3 respectively; The boundary map and the feature map are used together as input features of the multi-scale decoder. The input features are respectively passed through a 1×1 convolution layer, a 3×3 convolution layer and two 3×3 hole convolution layers to obtain four output results. Another 1×1 convolution layer integrates the four output results at the channel level, and then the features are further enhanced through the spatial attention module to obtain the overall features of the erosion gully image.

7. The multi-scale edge-aware erosion gully segmentation method according to claim 6, characterized in that: The output of a 1×1 convolutional layer is Y1=Conv1*1(X); The output of the 3×3 convolutional layer is Y2=Conv3*3(X); The output of the 3×3 dilated convolutional layer with a dilation rate of 2 is Y3=DilatedConv3*3(X,2); The output of the 3×3 dilated convolutional layer with a dilation rate of 3 is Y4=DilatedConv3*3(X,3); X represents the input feature, Conv1*1() represents a 1×1 convolution operation, Conv3*3(X) represents a 3×3 convolution operation, and DilatedConv3*3() represents a 3×3 dilated convolution operation.

8. The multi-scale edge-aware erosion gully segmentation method according to claim 1, characterized in that: The encoder is an encoder of the Segformer algorithm.

9. The multi-scale edge-aware erosion gully segmentation method according to claim 1, characterized in that: The method for obtaining the erosion gully image is as follows: Collecting remote sensing images of the target area through satellites, and cropping and scaling the remote sensing images in sequence so that the remote sensing images can adapt to the processing conditions of the semantic segmentation network; The adjusted remote sensing image is then standardized to obtain the erosion gully image.

10. The multi-scale edge-aware erosion gully segmentation method according to claim 9, characterized in that: The standardized expression is as follows: Among them, x′ represents the normalized pixel value, x i is the pixel value of the i-th pixel in the adjusted remote sensing image, i=1,2,...,N, N is the total number of pixels in the adjusted remote sensing image, The average value of all pixels in the adjusted remote sensing image.

Images