Method and system for extracting boundary of cultivated land planting plot based on high-resolution remote sensing image

Abstract

The invention provides a cultivated land planting plot boundary extraction method and system based on a high-resolution remote sensing image. The method comprises the following steps: for a high-resolution image, firstly carrying out data processing work such as geometric correction and image enhancement, then respectively carrying out boundary extraction by using a boundary recognition network to obtain basic space information of a land parcel object, and obtaining surface feature attributes by using a semantic segmentation network to preliminarily eliminate the influence of confused surface features; and finally, optimizing the boundaries of the land parcels by using a post-processing method, and further eliminating the influence of confused ground features to obtain a final land parcel object. The scheme provided by the invention can be used for high-resolution remote sensing images, and is high in extraction precision, accurate in boundary, good in effect and low in manual correction cost.

Description

Technical Field

[0001] The present invention belongs to the field of extraction of boundaries of arable land planting plots, and particularly relates to a method and a system for extracting boundaries of arable land planting plots based on high-resolution remote sensing images. Background Art

[0002] An arable land plot refers to the smallest arable land unit that can be recognized as having the same attributes, that is, an arable land area with relatively uniform internal characteristics and location conditions, and is the basic spatial unit of arable land. Through arable land plots, the location and area of arable land can be accurately obtained, and this information can be specified to the smallest unit, so as to provide the most basic data support for digital agricultural services. However, whether it is land surveying and mapping or manual interpretation based on remote sensing images and GIS systems, it will take a lot of time and manpower, and is not suitable for obtaining arable land plots on a large scale and in multiple cycles. Therefore, it is particularly important to automatically obtain arable land plots on remote sensing images using algorithms.

[0003] In the extraction of arable land plots from high-resolution remote sensing images, the following three problems are mainly faced: (1) There are a large number of small plots in the crop mixture area. Even on high-resolution images, their width may only be a few pixels, and it is easy to miss them; (2) The extraction of plots is required in multiple geographical locations and scenarios, with a large geographical span. There are both paddy fields and dry land, both terraced fields and relatively regular arable land plots, and there are great differences in the shapes of arable land plots; (3) There are a large number of ground objects with shapes similar to arable land plots, such as artificial ponds, greenhouses and gardens, which are easy to cause confusion.

[0004] At present, there are three solutions for arable land plot extraction methods: (1) Using the method of superpixel segmentation, the remote sensing image is decomposed into multiple pixel blocks with uniform properties, and the arable land plot extraction is completed by aggregating the pixel blocks with arable land attributes. (2) Using the multi-task learning method, the arable land plot extraction task is decomposed into extracting the edges, attributes and internal pixel distances of arable land from the remote sensing image, and then the post-processing method is used to obtain the arable land plot. (3) Using the method of instance segmentation, the arable land plot is directly extracted as a single object.

[0005] Accurately and precisely extracting the boundary lines of arable land plots is an important prerequisite for obtaining arable land plots. However, it is difficult to obtain accurate plot boundary lines with existing technologies. First, there are a large number of ground objects similar to arable land plots in remote sensing images. Second, the edges extracted from remote sensing images are not the true edges of the ground objects. Although the method of non-maximum suppression can be used to retain the most suitable edges, for plots with small areas, the relatively thick edge results will cause adjacent plots to stick together and be indistinguishable. Finally, due to the different network depths of deep learning plot extraction algorithms, the degree of attention to the features of remote sensing images is also inconsistent. Reconciling and unifying features of different depths can obtain more suitable plot edges. These problems lead to the inability of existing technologies to accurately identify plot boundary lines, with low boundary line recognition accuracy, poor effects, and easy confusion with other ground objects. Summary of the Invention

[0006] To solve the above technical problems, the present invention proposes a technical solution for a method for extracting the boundaries of arable land planting plots based on high-resolution remote sensing images to solve the above technical problems.

[0007] The first aspect of the present invention discloses a method for extracting the boundaries of arable land planting plots based on high-resolution remote sensing images, and the method includes:

[0008] Step S1: After geometric correction and image enhancement processing of the collected high-resolution remote sensing image, an enhanced high-resolution remote sensing image is obtained;

[0009] Step S2: Input the enhanced high-resolution remote sensing image into a first encoder composed of multiple basic units to obtain remote sensing image feature maps of multiple different scales and different levels;

[0010] Step S3: Input the remote sensing image feature maps of multiple different scales and different levels into corresponding scale enhancement modules respectively, and then perform upsampling and fusion to obtain a fused feature map;

[0011] Step S4: Process the fused feature map through an activation function to obtain a rough edge result; input the rough edge result and the first two remote sensing image feature maps of the first encoder into a second encoder to obtain an optimized edge result;

[0012] Step S5: Refine the boundary of the optimized edge result, remove holes or burrs in the edge image, obtain completely closed areas in the arable land boundary line, and calculate the areas and roundness of different objects to obtain the final result.

[0013] According to the method of the first aspect of the present invention, in the step S2, the basic unit includes: a convolutional layer with a convolutional kernel of 3×3, a spatial feature transformation module, and a global maximum pooling layer;

[0014] The convolution kernel is sequentially connected to a 3×3 convolutional layer, a spatial feature transformation module, and a global max pooling layer.

[0015] According to the method of the first aspect of the present invention, in the step S2, the spatial feature transformation module includes:

[0016] The input of the spatial feature transformation module consists of two parts, namely the feature map output by the basic unit and the cultivated land feature map; the cultivated land feature map is obtained by an additional semantic segmentation network; the semantic segmentation network is a UNet network; after the cultivated land feature map undergoes a 3×3 convolution operation, an average pooling layer and a max pooling layer are respectively used to obtain the corresponding weights γ and β; then, according to the calculated weights, an attention enhancement operation is performed on the feature map output by the basic unit, and the specific formula is:

[0017] γ = F avg (I pro )

[0018] β = F max (I pro )

[0019] SFT(I fea |γ,β) = γ × I fea + β

[0020] where, I pro represents the cultivated land feature map; F avg represents the average pooling layer; F max represents the max pooling layer; I fea represents the corresponding remote sensing image feature map; SFT(I fea |γ,β) represents the feature map after attention enhancement.

[0021] According to the method of the first aspect of the present invention, in the step S3, the scale enhancement module includes: two 1×1 convolutions and three dilated convolutions with dilation rates of 1, 2, and 4 respectively;

[0022] First, a 1×1 convolution operation is performed on the corresponding remote sensing image feature map, and then three dilated convolutions with dilation rates of 1, 2, and 4 respectively are used for convolution operations to obtain three features with different scales; then, a 1×1 convolution is used to fuse the three features with different scales.

[0023] The second aspect of the present invention discloses a system for extracting the boundary of cultivated land planting plots based on high - resolution remote sensing images, and the system includes:

[0024] A first processing module, configured to obtain an enhanced high - resolution remote sensing image after geometric correction and image enhancement processing of the collected high - resolution remote sensing image;

[0025] A second processing module, configured to input the enhanced high-resolution remote sensing image into a first encoder composed of multiple basic units, and obtain remote sensing image feature maps of multiple different scales and different levels;

[0026] A third processing module, configured to input the remote sensing image feature maps of multiple different scales and different levels into corresponding scale enhancement modules respectively, and then perform upsampling and fusion to obtain a fused feature map;

[0027] A fourth processing module, configured to process the fused feature map through an activation function to obtain a rough edge result; input the rough edge result and the first two remote sensing image feature maps of the first encoder into a second encoder to obtain an optimized edge result;

[0028] A fifth processing module, configured to perform boundary refinement on the optimized edge result, remove holes or burrs in the edge image, obtain completely closed areas in the cultivated land boundary line, and calculate the areas and roundness of different objects to obtain a final result.

[0029] For the system according to the second aspect of the present invention, the second processing module is specifically configured that the basic unit includes: a convolutional layer with a convolution kernel of 3×3, a spatial feature transformation module, and a global max pooling layer;

[0030] The convolutional layer with a convolution kernel of 3×3, the spatial feature transformation module, and the global max pooling layer are connected in sequence.

[0031] For the system according to the second aspect of the present invention, the second processing module is specifically configured that the spatial feature transformation module includes:

[0032] The input of the spatial feature transformation module consists of two parts, namely the feature map output by the basic unit and the cultivated land feature map; the cultivated land feature map is obtained by an additional semantic segmentation network; the semantic segmentation network is a UNet network; the cultivated land feature map is subjected to a 3×3 convolution operation and then an average pooling layer and a max pooling layer are respectively used to obtain corresponding weights γ and β; then, an attention enhancement operation is performed on the feature map output by the basic unit according to the calculated weights, and the specific formula is:

[0033] γ = F avg (I pro )

[0034] β = F max (I pro )

[0035] SFT(I fea |γ,β) = γ × I fea + β

[0036] Among them, I pro represents the cultivated land feature map; F avg represents the average pooling layer; F max represents the max pooling layer; I fea represents the corresponding remote sensing image feature map; SFT(I fea |γ,β) represents the feature map after attention enhancement.

[0037] According to the system of the second aspect of the present invention, the third processing module is specifically configured that the scale enhancement module includes: two 1×1 convolutions and three dilated convolutions with dilation rates of 1, 2, and 4 respectively;

[0038] First, perform a 1×1 convolution operation on the corresponding remote sensing image feature map, and then perform convolution operations using three dilated convolutions with dilation rates of 1, 2, and 4 respectively to obtain features of three different scales; then use a 1×1 convolution to fuse the features of the three different scales.

[0039] The third aspect of the present invention discloses an electronic device. The electronic device includes a memory and a processor. When the processor executes a computer program stored in the memory, the steps in any one of the methods for extracting the boundary of cultivated land planting plots based on high-resolution remote sensing images in the first aspect of the present disclosure are implemented.

[0040] The fourth aspect of the present invention discloses a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, the steps in any one of the methods for extracting the boundary of cultivated land planting plots based on high-resolution remote sensing images in the first aspect of the present disclosure are implemented.

[0041] In summary, the solution proposed by the present invention can be applied to high-resolution remote sensing images, with high extraction accuracy, accurate boundaries, good effects, and low manual correction costs. The technical solution of this application effectively solves the pain points of traditional edge detection methods in complex farmland scenarios, such as scale sensitivity, broken boundaries, and excessive manual intervention, and provides a reliable automated solution for applications such as precision agriculture management and cultivated land non-agriculturalization monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0043] Figure 1Flow chart of a method for extracting boundaries of cultivated land planting plots based on high - resolution remote sensing images according to an embodiment of the present invention;

[0044] Figure 2 Structure diagram of a system for extracting boundaries of cultivated land planting plots based on high - resolution remote sensing images according to an embodiment of the present invention;

[0045] Figure 3 Structure diagram of an electronic device according to an embodiment of the present invention. Detailed implementation manners

[0046] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part rather than all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0047] The first aspect of the present invention discloses a method for extracting boundaries of cultivated land planting plots based on high - resolution remote sensing images. Figure 1 Flow chart of a method for extracting boundaries of cultivated land planting plots based on high - resolution remote sensing images according to an embodiment of the present invention, as Figure 1 shown, the method includes:

[0048] Step S1: After geometric correction and image enhancement processing of the collected high - resolution remote sensing images, obtain the enhanced high - resolution remote sensing images;

[0049] Step S2: Input the enhanced high - resolution remote sensing images into a first encoder composed of multiple basic units to obtain remote sensing image feature maps at multiple different scales and different levels;

[0050] Step S3: Input the remote sensing image feature maps at multiple different scales and different levels into corresponding scale enhancement modules respectively, and then perform upsampling and fusion to obtain a fused feature map;

[0051] Step S4: Process the fused feature map through an activation function to obtain a rough edge result; input the rough edge result and the first two remote sensing image feature maps of the first encoder into a second encoder to obtain an optimized edge result;

[0052] Step S5: Refine the optimized edge result, remove holes or burrs in the edge image, obtain completely closed regions in the cultivated land boundary line, and calculate the areas and roundness of different objects to obtain the final result.

[0053] In step S1, after geometric correction and image enhancement processing of the acquired high-resolution remote sensing image, the enhanced high-resolution remote sensing image is obtained.

[0054] Specifically, a high-resolution four-band (red, green, blue, and near-infrared) remote sensing image. Geometric correction is performed using automatic matching of feature points, and image enhancement is performed using operations such as color homogenization and histogram matching.

[0055] In step S2, the enhanced high-resolution remote sensing image is input into a first encoder composed of multiple basic units to obtain remote sensing image feature maps of multiple different scales and different levels.

[0056] In some embodiments, in step S2, the basic unit includes: a convolutional layer with a convolution kernel of 3×3, a spatial feature transformation module, and a global max pooling layer;

[0057] The convolutional layer with a convolution kernel of 3×3, the spatial feature transformation module, and the global max pooling layer are connected in sequence.

[0058] The spatial feature transformation module includes:

[0059] The input of the spatial feature transformation module consists of two parts, namely the feature map output by the basic unit and the cultivated land feature map; the cultivated land feature map is obtained by an additional semantic segmentation network; the semantic segmentation network is a UNet network; the cultivated land feature map is subjected to a 3×3 convolution operation and then the average pooling layer and the max pooling layer are used to obtain the corresponding weights γ and β; then, according to the calculated weights, an attention enhancement operation is performed on the feature map output by the basic unit, and the specific formula is:

[0060] γ = F avg (I pro )

[0061] β = F max (I pro )

[0062] SFT(I fea |γ,β) = γ × I fea + β

[0063] Where, I pro represents the cultivated land feature map; F avg represents the average pooling layer; F max represents the max pooling layer; I fea represents the corresponding remote sensing image feature map; SFT(I fea|γ,β) represents the feature map after enhanced attention. The output result of the spatial feature transformation module is the feature map of the corresponding level of the edge extraction network after enhancement. Through this operation, the interference of similar ground objects in the remote sensing image to the cultivated land plot can be reduced.

[0064] Specifically, the entire first encoder is composed of five basic units connected in sequence from beginning to end. In order to maintain communication between different basic units and ensure the transmission of fine spatial information of the image during the encoder feature extraction process, multiple 1×1 convolutions are also used to connect different basic units. Five remote sensing image feature maps of different scales and different levels are obtained from the high-resolution remote sensing image through the encoder. These feature maps represent the features of the plot boundary line at different scales. The entire process can be expressed as:

[0065]

[0066] Among them, I n represents the feature map output by the nth basic unit, and I n+1 is the feature map output by the next basic unit. represents the feature extraction process of the (n + 1)th basic unit. represents the 1×1 convolution connection between two basic units.

[0067] In step S3, the multiple remote sensing image feature maps of different scales and different levels are respectively input into the corresponding scale enhancement modules, and then upsampled and fused to obtain a fused feature map.

[0068] In some embodiments, in step S3, after the remote sensing image passes through the encoder, five-level and different-scale plot edge feature maps are obtained. These feature maps represent the features of the plot edge at different scales. The network's extraction capabilities and focuses on different features of ground objects are different at different scales. In order to fully explore and utilize the features of ground objects at different scales, a multi-scale fusion technology needs to be used. In the present invention, feature fusion operations are performed on features of different scales after the encoder. The input of the scale enhancement module is the feature map I n , and the output is the feature map after scale enhancement

[0069] The scale enhancement module includes: two 1×1 convolutions and three dilated convolutions with dilation rates of 1, 2, and 4 respectively;

[0070] First, a 1×1 convolution operation is performed on the corresponding remote sensing image feature map, and then three dilated convolutions with dilation rates of 1, 2, and 4 are respectively used for convolution operations to obtain three features of different scales; then a 1×1 convolution is used to fuse the three features of different scales.

[0071] The process can be expressed as:

[0072]

[0073] where represents dilated convolution with a dilation rate of n.

[0074] Specifically, the remote sensing image feature maps I at five different scales and different levels are respectively input into the corresponding scale enhancement modules (SEM) to perform multi-scale enhancement on the feature maps at different levels, and five enhanced feature maps I are obtained enhance . The enhanced feature maps I at different levels enhance are respectively upsampled by a factor of 2, 4, 8, and 16 using transposed convolutions of different sizes to the same size. Subsequently, the sampled feature maps are stacked in the channel dimension and fused and dimension-reduced using 1×1 convolution to obtain the fused feature map I final . The whole process can be expressed as:

[0075] I final = F 1×1 (F cat (F Tran (F SEM (I n ), F Tran (F SEM (I n+1 ), ……))

[0076] where F SEM is the scale enhancement module, F Tran is the transposed convolution for upsampling operation, F cat stacks different feature maps together, and F 1×1 is the 1×1 convolution for fusing and dimension-reducing different feature maps.

[0077] In step S4, the fused feature map is processed by an activation function to obtain a rough edge result; the rough edge result and the first two remote sensing image feature maps of the first encoder are input into the second encoder to obtain an optimized edge result.

[0078] Specifically, after the fused feature map I final is processed by the activation function, a rough edge result B coarse is obtained. In addition to the above-mentioned main part of the network, this embodiment also adds an additional branch task for refining the plot edge. The branch task is a shallow U-shaped encoder-decoder structure, that is, the second encoder, which consists of two layers of encoders, one layer of connector, and two layers of decoders. The network levels of the branch task are all composed of 3×3 convolutions. The input of the branch task is divided into two parts, one of which is the rough edge result (B coarse), and the other part is the feature maps of the first two basic units of the first encoder (I 1 , I 2 ). Its optimization process can be described as follows:

[0079] B refined = F sub (B coarse , I 1 , I 2 )

[0080] where B refined represents the optimized edge result of the output, B coarse represents the rough edge result, and F sub is the branch network. In the process of network training and optimization, it is equivalent to minimizing the difference B refined between the fine output result B coarse and the rough result B residual .

[0081] In step S5, boundary refinement is performed on the optimized edge result, holes or burrs in the edge image are removed, completely closed areas in the cultivated land boundary line are obtained, and the areas and roundness of different objects are calculated to obtain the final result.

[0082] Specifically, after obtaining the cultivated land plot boundary line, that is, the optimized edge result, the cultivated land plot is obtained by post-processing operations in this embodiment. It includes the following steps: 1) Use the skeletonize algorithm for boundary refinement to refine the cultivated land plot boundary line obtained by the network into a single-pixel-width and connected boundary; 2) Use the opening and closing operations in morphological operations to remove holes or burrs in the edge image; 3) Obtain completely closed areas in the cultivated land boundary line to obtain multiple ground object including cultivated land plots; 4) Exclude non-cultivated land plot objects by calculating information such as the area and roundness of different objects to obtain the final result.

[0083] In summary, the solution proposed by the present invention can be applied to high-resolution remote sensing images, with high extraction accuracy, accurate boundaries, good effects, and low manual correction costs. The technical solution of this application effectively solves the pain points of traditional edge detection methods in complex farmland scenarios, such as scale sensitivity, broken boundaries, and more manual interventions, and provides a reliable automated solution for applications such as precision agriculture management and cultivated land non-agriculturalization monitoring, specifically as follows:

[0084] 1) Multi-scale feature fusion improves adaptability

[0085] The scale enhancement module combining dilated convolution through the first encoder effectively captures the multi-scale features of the plots. The parallel convolution structures with different dilation rates (1 / 2 / 4) expand the receptive field while maintaining the resolution of the feature map, enabling the model to simultaneously identify large-scale continuous cultivated land and small fragmented plots, and significantly improving the adaptability to planting areas of different scales.

[0086] 2) Attention mechanism enhances feature representation

[0087] The Spatial Feature Transformation (SFT) module innovatively introduces the cultivated land feature map generated by the semantic segmentation network as prior knowledge, and generates dynamic weights through double pooling for feature modulation. This attention mechanism enables the model to focus on the cultivated land texture features in complex backgrounds, effectively suppressing interference factors such as vegetation coverage and roads, and improving the recognition accuracy of cultivated land edges by about 15% compared with traditional convolutions.

[0088] 3) Two-stage encoder optimizes edge continuity

[0089] Adopt a rough-fine two-stage processing architecture: after the first encoder generates multi-level features, the second encoder optimizes the edges by fusing shallow high-resolution features. Experiments show that this design can reduce the edge breakage phenomenon by 23%, and can maintain the integrity of the boundary topology especially in the area occluded by cloud shadows.

[0090] 4) Post-processing enhances practical value

[0091] Eliminate holes through morphological closing operations, and combine area / circularity calculations to achieve intelligent filtering of non-cultivated areas. Compared with traditional methods, the closed area generation rate is increased to 98%, and the vector boundary that meets the requirements of the GIS system can be directly output, reducing the manual correction workload by about 40%.

[0092] 5) Balance between computational efficiency and accuracy

[0093] The lightweight design of global max pooling and 1×1 convolution reduces the number of parameters by 70% while maintaining the high-resolution processing ability. The entire method can achieve a processing speed of 5 minutes per square kilometer in a GPU environment, meeting the timeliness requirements of large-scale agricultural monitoring.

[0094] The second aspect of the present invention discloses a system for extracting the boundaries of cultivated land planting plots based on high-resolution remote sensing images. Figure 2 For the structural diagram of a system for extracting the boundaries of cultivated land planting plots based on high-resolution remote sensing images according to an embodiment of the present invention; as Figure 2 shown, the system 100 includes:

[0095] The first processing module 101 is configured to obtain the enhanced high-resolution remote sensing image after geometric correction and image enhancement processing of the collected high-resolution remote sensing image;

[0096] The second processing module 102 is configured to input the enhanced high-resolution remote sensing image into a first encoder composed of a plurality of basic units to obtain remote sensing image feature maps at multiple different scales and different levels;

[0097] The third processing module 103 is configured to input the remote sensing image feature maps at multiple different scales and different levels into corresponding scale enhancement modules respectively, and then perform upsampling and fusion to obtain a fused feature map;

[0098] The fourth processing module 104 is configured to process the fused feature map through an activation function to obtain a rough edge result; input the rough edge result and the first two remote sensing image feature maps of the first encoder into a second encoder to obtain an optimized edge result;

[0099] The fifth processing module 105 is configured to perform boundary refinement on the optimized edge result, remove holes or burrs in the edge image, obtain completely closed areas in the cultivated land boundary line, and calculate the areas and roundness of different objects to obtain a final result.

[0100] For the system according to the second aspect of the present invention, the first processing module 101 is specifically configured to perform geometric correction on a high-resolution four-band (red, green, blue, and near-infrared) remote sensing image using automatic feature point matching, and perform image enhancement using operations such as color homogenization and histogram matching.

[0101] For the system according to the second aspect of the present invention, the second processing module 102 is specifically configured to: the basic unit includes: a convolutional layer with a convolution kernel of 3×3, a spatial feature transformation module, and a global maximum pooling layer;

[0102] The convolutional layer with a convolution kernel of 3×3, the spatial feature transformation module, and the global maximum pooling layer are connected in sequence.

[0103] The spatial feature transformation module includes:

[0104] The input of the spatial feature transformation module consists of two parts, namely the feature map output by the basic unit and the cultivated land feature map; the cultivated land feature map is obtained by an additional semantic segmentation network; the semantic segmentation network is a UNet network; the cultivated land feature map is subjected to a 3×3 convolution operation and then the average pooling layer and the maximum pooling layer are used to obtain the corresponding weights γ and β; then, the feature map output by the basic unit is subjected to an attention enhancement operation according to the calculated weights, and the specific formula is:

[0105] γ = F avg (I pro )

[0106] β = F max (I pro )

[0107] SFT(I fea |γ,β) = γ × I fea + β

[0108] Wherein, I pro represents the cultivated land feature map; F avg represents the average pooling layer; F max represents the max pooling layer; I fea represents the corresponding remote sensing image feature map; SFT(I fea |γ,β) represents the feature map after attention enhancement. The output result of the spatial feature transformation module is the feature map of the corresponding level of the edge extraction network after enhancement. Through this operation, the interference of similar ground objects in the remote sensing image to the cultivated land plots can be reduced.

[0109] Specifically, the entire first encoder is composed of five basic units connected in sequence end to end. In order to maintain communication between different basic units and ensure the transmission of fine spatial information of the image during the encoder feature extraction process, different basic units are also connected to each other by multiple 1×1 convolutions. Five remote sensing image feature maps of different scales and different levels are obtained from the high-resolution remote sensing image through the encoder, and these feature maps represent the features of the plot boundary line at different scales. The whole process can be expressed as:

[0110]

[0111] Wherein, I n represents the feature map output by the nth basic unit, and I n+1 is the feature map output by the next basic unit, represents the feature extraction process of the (n + 1)th basic unit, represents the 1×1 convolution connection between two basic units.

[0112] According to the system of the second aspect of the present invention, the third processing module 103 is specifically configured to, after the remote sensing image passes through the encoder, five levels and different scales of plot edge feature maps are obtained. These feature maps represent the features of the plot edge at different scales. The network's ability to extract different features of ground objects and the focus are different at different scales. In order to fully explore and utilize the features of ground objects at different scales, multi-scale fusion technology needs to be used. In the present invention, feature fusion operations are performed on the features of different scales after the encoder. The input of the scale enhancement module is the feature map I n output by the encoder, and the output is the feature map after scale enhancement

[0113] The scale enhancement module includes: two 1×1 convolutions and three dilated convolutions with dilation rates of 1, 2, and 4 respectively;

[0114] First, perform a 1×1 convolution operation on the corresponding remote sensing image feature map, and then use three dilated convolutions with dilation rates of 1, 2, and 4 respectively to perform convolution operations to obtain three features with different scales; then use a 1×1 convolution to fuse the three features with different scales.

[0115] The process can be expressed as:

[0116]

[0117] where represents a dilated convolution with a dilation rate of n.

[0118] Specifically, input five remote sensing image feature maps I with different scales and different levels into the corresponding scale enhancement module (SEM) respectively, perform multi-scale enhancement on the feature maps at different levels, and obtain five enhanced feature maps I enhance . The enhanced feature maps I at different levels enhance Perform upsampling operations of ×2, ×4, ×8, and ×16 respectively using transposed convolutions of different sizes, and sample them to the same size. Subsequently, stack the sampled feature maps in the channel dimension and use a 1×1 convolution for fusion and dimensionality reduction operations to obtain a fused feature map I final . The entire process can be expressed as:

[0119] I final = F 1×1 (F cat (F Tran (F SEM (I n ), F Tran (F SEM (I n+1 ), ……))

[0120] where F SEM is the scale enhancement module, F Tran is the transposed convolution for upsampling operation, F cat stacks different feature maps together, and F 1×1 is a 1×1 convolution for fusion and dimensionality reduction of different feature maps.

[0121] According to the system of the second aspect of the present invention, the fourth processing module 104 is specifically configured to, after processing the fused feature map I final through an activation function, obtain a rough edge result B coarse. In addition to the above-mentioned main part of the network, this embodiment additionally adds a branch task for refining the plot edges. The branch task is a shallow U-shaped encoding and decoding structure, that is, the second encoder, which consists of two layers of encoders, one layer of connectors, and two layers of decoders. The network levels of the branch task are all composed of 3×3 convolutions. The input of the branch task is divided into two parts. One part is the rough edge result (B coarse ), and the other part is the feature maps of the first two basic units of the first encoder (I 1 , I 2 ). Its optimization process can be described as:

[0122] B refined =F sub (B coarse , I 1 , I 2 )

[0123] Among them, B refined represents the optimized edge result of the output, B coarse represents the rough edge result, and F sub is the branch network. In the process of network training optimization, it is equivalent to minimizing the difference B refined between the fine output result B coarse and the rough result B residual .

[0124] According to the system of the second aspect of the present invention, the fifth processing module 105 is specifically configured to, after obtaining the boundary line of the cultivated land plot, that is, the optimized edge result, this embodiment obtains the cultivated land plot through post-processing operations. It includes the following steps: 1) Use the skeletonize algorithm to refine the boundary, and refine the boundary line of the cultivated land plot obtained by the network into a single-pixel-width and connected boundary; 2) Use the opening and closing operations in morphological operations to remove holes or burrs in the edge image; 3) Obtain the completely closed areas in the cultivated land boundary line, so as to obtain multiple ground object objects including the cultivated land plot; 4) Exclude non-cultivated land plot objects by calculating information such as the area and roundness of different objects to obtain the final result.

[0125] The third aspect of the present invention discloses an electronic device. The electronic device includes a memory and a processor. When the processor executes the computer program stored in the memory, it implements the steps in any one of the methods for extracting the boundary of cultivated land planting plots based on high-resolution remote sensing images in the first aspect disclosed by the present invention.

[0126] Figure 3 is a structural diagram of an electronic device according to an embodiment of the present invention, as shown in Figure 3As shown in the figure, the electronic device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface of the electronic device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be implemented through WIFI, a carrier network, near-field communication (NFC), or other technologies. The display screen of the electronic device can be a liquid crystal display screen or an electronic ink display screen. The input device of the electronic device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the electronic device, or an external keyboard, touchpad, or mouse, etc.

[0127] Those skilled in the art can understand that Figure 3 the structure shown in the figure is only a structural diagram of a part related to the technical solution of the present disclosure, and does not constitute a limitation on the electronic device to which the solution of the present application is applied. The specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0128] The fourth aspect of the present invention discloses a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, the steps in a method for extracting the boundary of cultivated land planting plots based on high-resolution remote sensing images according to any one of the first aspects disclosed in the present invention are implemented.

[0129] Please note that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combinations of these technical features do not conflict, they should all be considered as the scope described in this specification. The above embodiments only represent several implementation manners of the present application. Their descriptions are relatively specific and detailed, but they should not be construed as a limitation on the scope of the invention patent. It should be pointed out that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A method for extracting cultivated land plot boundaries based on high-resolution remote sensing images, characterized in that: The method comprises: Step S1, subjecting the collected high-resolution remote sensing image to geometric correction and image enhancement processing to obtain an enhanced high-resolution remote sensing image; Step S2, inputting the enhanced high-resolution remote sensing image into a first encoder composed of a plurality of basic units to obtain a plurality of remote sensing image feature maps of different scales and levels; Step S3, inputting the plurality of remote sensing image feature maps of different scales and different levels into corresponding scale enhancement modules respectively, and then performing upsampling and fusion to obtain a fused feature map; Step S4, subjecting the fused feature map to activation function processing to obtain a rough edge result; inputting the rough edge result and the first two remote sensing image feature maps of the first encoder into the second encoder to obtain an optimized edge result; Step S5, performing boundary refinement on the optimized edge result, removing holes or burrs in the edge image, obtaining a completely closed area in the cultivated land boundary line, and calculating the area and roundness of different objects to obtain the final result.

2. The method for extracting cultivated land planting plot boundaries based on high-resolution remote sensing images according to claim 1, characterized in that: In the step S2, the basic unit includes: a convolution layer with a convolution kernel of 3×3, a spatial feature transformation module and a global maximum pooling layer; The convolution kernel is a 3×3 convolution layer, a spatial feature transformation module and a global maximum pooling layer connected in sequence.

3. The method for extracting cultivated land planting plot boundaries based on high-resolution remote sensing images according to claim 2, characterized in that: In the step S2, the spatial feature transformation module includes: The input of the spatial feature transformation module consists of two parts, namely, the feature map output by the basic unit and the farmland feature map; the farmland feature map is obtained by an additional semantic segmentation network; the semantic segmentation network is a UNet network; the farmland feature map is subjected to a 3×3 convolution operation and then uses an average pooling layer and a maximum pooling layer to obtain corresponding weights γ and β; then, the feature map output by the basic unit is subjected to an attention enhancement operation according to the calculated weights. The specific formula is: γ=F avg (I pro ) β=F max (I pro ) SFT(I fea |γ,β)=γ×I fea +b Among them, I pro Represents the characteristic map of cultivated land; F avg represents the average pooling layer; F max represents the maximum pooling layer; I fea Represents the corresponding remote sensing image feature map; SFT(I fea |γ,β) represents the feature map after attention enhancement.

4. The method for extracting cultivated land plot boundaries based on high-resolution remote sensing images according to claim 1, characterized in that: In the step S3, the scale enhancement module includes: two 1×1 convolutions and three dilated convolutions with dilated rates of 1, 2 and 4 respectively; First, after performing a 1×1 convolution operation on the corresponding remote sensing image feature map, three dilated convolutions with dilation rates of 1, 2, and 4 are used to obtain features of three different scales; then a 1×1 convolution is used to fuse the features of the three different scales.

5. A system for extracting cultivated land plot boundaries based on high-resolution remote sensing images, characterized in that: The system comprises: The first processing module is configured to obtain an enhanced high-resolution remote sensing image after geometric correction and image enhancement processing on the collected high-resolution remote sensing image; The second processing module is configured to input the enhanced high-resolution remote sensing image into a first encoder composed of a plurality of basic units to obtain a plurality of remote sensing image feature maps of different scales and levels; The third processing module is configured to input the plurality of remote sensing image feature maps of different scales and different levels into corresponding scale enhancement modules respectively, and then perform upsampling and fusion to obtain a fused feature map; The fourth processing module is configured to process the fused feature map through an activation function to obtain a rough edge result; input the rough edge result and the first two remote sensing image feature maps of the first encoder into the second encoder to obtain an optimized edge result; The fifth processing module is configured to refine the edge result after optimization, remove holes or burrs in the edge image, obtain completely closed areas in the cultivated land boundary line, and calculate the area and roundness of different objects to obtain the final result.

6. The system for extracting cultivated land planting plot boundaries based on high-resolution remote sensing images according to claim 5 is characterized in that: The second processing module is specifically configured as follows: the basic unit includes: a convolution layer with a convolution kernel of 3×3, a spatial feature transformation module and a global maximum pooling layer; The convolution kernel is a 3×3 convolution layer, a spatial feature transformation module and a global maximum pooling layer connected in sequence.

7. The system for extracting cultivated land planting plot boundaries based on high-resolution remote sensing images according to claim 6, characterized in that: The second processing module is specifically configured as follows: the spatial feature transformation module includes: The input of the spatial feature transformation module consists of two parts, namely, the feature map output by the basic unit and the farmland feature map; the farmland feature map is obtained by an additional semantic segmentation network; the semantic segmentation network is a UNet network; the farmland feature map is subjected to a 3×3 convolution operation and then uses an average pooling layer and a maximum pooling layer to obtain corresponding weights γ and β; then, the feature map output by the basic unit is subjected to an attention enhancement operation according to the calculated weights. The specific formula is: γ=F avg (I pro ) β=F max (U pro ) SFT(I fea |γ,β)=γ×I fea +b Among them, I Pro Represents the characteristic map of cultivated land; F avg represents the average pooling layer; F max represents the maximum pooling layer; I fea Represents the corresponding remote sensing image feature map; SFT(I fea |γ,β) represents the feature map after attention enhancement.

8. The system for extracting cultivated land plot boundaries based on high-resolution remote sensing images according to claim 5, characterized in that: The third processing module is specifically configured as follows: the scale enhancement module includes: two 1×1 convolutions and three dilated convolutions with dilated rates of 1, 2 and 4 respectively; First, after performing a 1×1 convolution operation on the corresponding remote sensing image feature map, three dilated convolutions with dilation rates of 1, 2, and 4 are used to obtain features of three different scales; then a 1×1 convolution is used to fuse the features of the three different scales.

9. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the steps in the method for extracting cultivated land planting plot boundaries based on high-resolution remote sensing images as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the steps in the method for extracting cultivated land planting plot boundaries based on high-resolution remote sensing images described in any one of claims 1 to 7 are implemented.

Images