Multi-source remote sensing image-based peasant household land checking method and device, storage medium and processor

Abstract

The invention discloses a peasant household land checking method and device based on a multi-source remote sensing image, a storage medium and a processor. In the scheme, a multi-source remote sensing image of a target farmer is obtained, the multi-source remote sensing image comprises a multi-spectral remote sensing image and a panchromatic remote sensing image, firstly, based on the multi-spectral remote sensing image, a semantic segmentation technology is adopted to identify a regional crop planting condition, and a crop distribution result of a land of the target farmer is obtained; and then based on the crop distribution result of the target farmer land and the panchromatic remote sensing image, a semantic segmentation technology is adopted to extract the farmer land parcel, and a parcel-level crop distribution result of the target farmer land is obtained. Compared with the prior art that a single remote sensing image is used, the recognition precision of the farmer land area is not enough, and richer information such as crop planting types cannot be obtained, the multi-source remote sensing image is segmented through the semantic segmentation technology, and the accuracy of the obtained farmer land information is improved. Compared with the prior art, the method has obvious advantages.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a farmer's land verification method, device, storage medium and processor based on multi-source remote sensing images. Background Art

[0002] Since farmers' planting scale is small and scattered, it is difficult to obtain real information about farmers' land assets. Remote sensing is a non-contact, long-distance detection technology. It uses visible light, infrared, microwave and other detection instruments on satellites to detect and identify ground materials. It has the characteristics of long distance and wide coverage. Remote sensing technology can be used to obtain ground satellite images to analyze farmers' farmland information.

[0003] The existing farmer land verification method based on remote sensing technology is to extract farmland by calculating the vegetation index of remote sensing images and adding the binary segmentation threshold algorithm OSTU. However, this method uses relatively simple spectral information of ground objects and can only identify the area of ​​farmland. The identification accuracy of farmland is not enough, and it is impossible to obtain richer information such as farmland planting type and planting detail area.

[0004] Therefore, how to use remote sensing technology to improve the accuracy of farmers' land information is a technical problem that needs to be solved urgently. Summary of the invention

[0005] Based on the above problems, the present application provides a farmer's land verification method, device, storage medium and processor based on multi-source remote sensing images, with the aim of improving the accuracy of the acquired farmer's land information.

[0006] The embodiments of the present application disclose the following technical solutions:

[0007] The first aspect of the present application provides a farmer's land verification method based on multi-source remote sensing images, the method comprising:

[0008] Acquire multi-source remote sensing images of the target farmer's land; the multi-source remote sensing images include multi-spectral remote sensing images and panchromatic remote sensing images;

[0009] Performing standardization processing on the multispectral remote sensing image and the panchromatic remote sensing image;

[0010] Inputting the standardized multispectral remote sensing image into a first semantic segmentation network, the first semantic segmentation network performs semantic segmentation based on the standardized multispectral remote sensing image, and outputs the crop distribution result of the target farmer's land; the first semantic segmentation network is obtained after training based on the multispectral remote sensing image training samples;

[0011] The standardized panchromatic remote sensing image is input into a second semantic segmentation network, and the second semantic segmentation network performs semantic segmentation according to the crop distribution result of the target farmer's land and the standardized panchromatic remote sensing image, and outputs the plot-level crop distribution result of the target farmer's land; the second semantic segmentation network is obtained after training based on the panchromatic remote sensing image training samples;

[0012] Compare the plot-level crop distribution results of the target farmer's land with the target farmer's enclosure data to verify the target farmer's land assets.

[0013] In an optional implementation, the structures of the first semantic segmentation network and the second semantic segmentation network both include:

[0014] An encoder, a decoder and a jump connection; a spatial attention mechanism is added to the encoder and the decoder respectively; the spatial attention mechanism highlights the local features of the space by dynamically adjusting the feature weights.

[0015] In an optional implementation, the first semantic segmentation network is trained by the following steps:

[0016] Acquire a multispectral remote sensing image training sample corresponding to the first semantic segmentation network; the multispectral remote sensing image training sample includes a multispectral remote sensing image sample map and a multispectral remote sensing image label;

[0017] The multispectral remote sensing image training samples are subjected to standardization processing; the standardization processing includes standardization processing of the multispectral remote sensing image sample map and standardization processing of the multispectral remote sensing image labels; the standardization processing of the multispectral remote sensing image labels includes resampling, geographic registration and cutting and blocking;

[0018] Based on the standardized multispectral remote sensing image training samples, the network parameters of the first semantic segmentation network are determined through training and testing on the initial network structure of the first semantic segmentation network.

[0019] In an optional implementation, the second semantic segmentation network is trained by the following steps, including:

[0020] Obtaining a panchromatic remote sensing image training sample corresponding to the second semantic segmentation network; the panchromatic remote sensing image training sample includes a panchromatic remote sensing image and a panchromatic remote sensing image label;

[0021] The panchromatic remote sensing image training samples are subjected to standardization processing; the standardization processing includes standardization processing of the panchromatic remote sensing image sample map and standardization processing of the panchromatic remote sensing image labels; the standardization processing of the panchromatic remote sensing image labels includes resampling, geographic registration and cutting and blocking;

[0022] Based on the standardized panchromatic remote sensing image training samples, the network parameters of the second semantic segmentation network are determined through training and testing on the initial network structure of the second semantic segmentation network.

[0023] In an optional implementation, the multispectral remote sensing image is subjected to standardization processing, including:

[0024] Performing atmospheric correction on the multispectral remote sensing image;

[0025] Resampling the multispectral remote sensing image after atmospheric correction to unify the resolution of the multispectral remote sensing image after atmospheric correction to a preset value;

[0026] Perform band fusion on the resampled multispectral remote sensing images and synthesize the data of multiple channels into the same multispectral remote sensing image for use;

[0027] The multispectral remote sensing image after band fusion is cut and divided into blocks according to the preset pixel size.

[0028] In an optional implementation, the panchromatic remote sensing image is subjected to standardization processing, including:

[0029] The full-color remote sensing image is cut into blocks according to the preset pixel size.

[0030] In an optional implementation, comparing the plot-level crop distribution result of the target farmer's land with the enclosure data of the target farmer to verify the land assets of the target farmer includes:

[0031] If the crop types analyzed by the crop distribution results match the crop types provided by the enclosure data, and the ratio of the difference between the farmland area analyzed by the crop distribution results and the farmland area provided by the enclosure data to the farmland area provided by the enclosure data is less than a preset ratio, a first verification result of the target farmer is generated; if the crop types analyzed by the crop distribution results do not match the crop types provided by the enclosure data, or the ratio of the difference between the farmland area analyzed by the crop distribution results and the farmland area provided by the enclosure data to the farmland area provided by the enclosure data is greater than or equal to a preset ratio, a second verification result of the target farmer is generated;

[0032] The first verification result indicates that the land assets of the target farmer have been automatically verified;

[0033] The second verification result indicates that a manual verification of the target farmer's land assets is required.

[0034] The second aspect of the present application provides a farmer's land verification device based on multi-source remote sensing images, the device comprising:

[0035] A remote sensing image acquisition module is used to acquire multi-source remote sensing images of target farmer's land; the multi-source remote sensing images include multi-spectral remote sensing images and panchromatic remote sensing images;

[0036] A remote sensing image standardization processing module, used for performing standardization processing on the multispectral remote sensing image and the panchromatic remote sensing image;

[0037] A remote sensing image semantic segmentation module is used to input a standardized multispectral remote sensing image into a first semantic segmentation network, and the first semantic segmentation network performs semantic segmentation based on the standardized multispectral remote sensing image, and outputs a crop distribution result of a target farmer's land; the first semantic segmentation network is obtained based on a multispectral remote sensing image training sample; the standardized panchromatic remote sensing image is input into a second semantic segmentation network, and the second semantic segmentation network performs semantic segmentation based on the crop distribution result of the target farmer's land and the standardized panchromatic remote sensing image, and outputs a plot-level crop distribution result of the target farmer's land; the second semantic segmentation network is obtained based on a panchromatic remote sensing image training sample;

[0038] The farmer land verification module is used to compare the plot-level crop distribution results of the target farmer's land with the target farmer's enclosure data to verify the target farmer's land assets.

[0039] The third aspect of the present application provides a computer-readable storage medium, in which a computer program is stored. When the program is executed by a processor, a farmer's land verification method based on multi-source remote sensing images as provided in any implementation of the first aspect is implemented.

[0040] A fourth aspect of the present application provides a processor for running a computer program, which, when running, executes a farmer's land verification method based on multi-source remote sensing images as provided in any implementation of the first aspect.

[0041] Compared with the prior art, this application has the following beneficial effects:

[0042] The present application provides a farmer's land verification method based on multi-source remote sensing images, which obtains multi-source remote sensing images of target farmers, wherein the multi-source remote sensing images include multispectral remote sensing images and panchromatic remote sensing images. First, based on the multispectral remote sensing images, remote sensing semantic segmentation technology is used to identify the regional crop planting conditions to obtain the crop distribution results of the target farmer's land. In order to make up for the lack of details in the multispectral remote sensing images, semantic segmentation technology is used to extract rural land parcels based on the crop distribution results of the target farmer's land and the panchromatic remote sensing images to obtain the plot-level crop distribution results of the target farmer's land. The plot-level crop distribution results of the target farmer's land are compared with the farmer's enclosure data to verify the authenticity of the farmer's land assets.

[0043] The present application provides a farmer's land verification method based on multi-source remote sensing images. Compared with the farmer's land verification method using a single remote sensing image, the method adopts semantic segmentation technology to improve the accuracy of the extracted farmer's farmland. In addition to obtaining the scope and area of ​​the farmer's land, the multispectral remote sensing image and the panchromatic remote sensing image can also obtain the farmer's crop planting information, so as to have a more comprehensive and accurate understanding of the farmer's land information and provide more effective help in verifying the authenticity of the farmer's land assets. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0045] Figure 1 A flowchart of a farmer's land verification method based on multi-source remote sensing images provided in an embodiment of the present application;

[0046] Figure 2 A flowchart of a farmer's land verification method based on multi-source remote sensing images provided in an embodiment of the present application;

[0047] Figure 3 The crop distribution result diagram provided by the embodiment of the present application;

[0048] Figure 4 A plot-level crop distribution result map provided in an embodiment of the present application;

[0049] Figure 5 A schematic diagram of the structure of a farmer's land verification device based on multi-source remote sensing images provided in an embodiment of the present application. DETAILED DESCRIPTION

[0050] As described above, the current farmer land verification method based on remote sensing technology is to extract farmland by calculating the vegetation index of remote sensing images and using the binary segmentation threshold algorithm OSTU. However, this method uses relatively simple spectral information of ground objects and can only identify the area of ​​farmland. The accuracy of farmland identification is not enough, and it is impossible to obtain richer information such as farmland planting type and planting detail area.

[0051] In view of the above problems, the inventors have proposed a farmer's land verification method, device, storage medium and processor based on multi-source remote sensing images after research. In this scheme, firstly, a multi-source remote sensing image of the target farmer's land is obtained; the multi-source remote sensing image includes a multispectral remote sensing image and a panchromatic remote sensing image; then, the multispectral remote sensing image and the panchromatic remote sensing image are standardized; the standardized multispectral remote sensing image is input into a first semantic segmentation network, and the first semantic segmentation network performs semantic segmentation according to the standardized multispectral remote sensing image, and outputs the crop distribution result of the target farmer's land; the first semantic segmentation network is obtained after training based on multispectral remote sensing image training samples; then, the standardized panchromatic remote sensing image is input into a second semantic segmentation network, and the second semantic segmentation network performs semantic segmentation according to the crop distribution result of the target farmer's land and the standardized panchromatic remote sensing image, and outputs the plot-level crop distribution result of the target farmer's land; the second semantic segmentation network is obtained after training based on the panchromatic remote sensing image training samples; finally, the plot-level crop distribution result of the target farmer's land is compared with the enclosure data of the target farmer to verify the land assets of the target farmer.

[0052] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0053] See also Figure 1 , which is a flow chart of a farmer's land verification method based on multi-source remote sensing images provided in an embodiment of the present application. Figure 1 As shown, the method comprises the following steps:

[0054] S101. Acquire multi-source remote sensing images of target farmer's land.

[0055] The multi-source remote sensing images include multispectral remote sensing images and panchromatic remote sensing images.

[0056] Multispectral remote sensing images refer to remote sensing images obtained by sensors with multiple different bands. Each band corresponds to a specific interval in the electromagnetic spectrum, usually including multiple bands such as visible light, near infrared, and mid-infrared, and is divided into multiple narrow bands for detection. Based on these spectral information, more detailed spectral characteristics of ground objects can be obtained to realize the application of crop identification.

[0057] Panchromatic remote sensing images refer to high-resolution remote sensing images acquired in a single band (usually the visible light band). Panchromatic images have a high spatial resolution, which can reach the sub-meter level. Because of their high resolution, they can clearly express the spatial structure and surface texture characteristics of ground objects, distinguish the finer components inside the ground objects, and make the edge information of the ground objects clearer, and can clearly display surface details such as buildings, roads, trees, etc.

[0058] S102: performing standardization processing on the multispectral remote sensing image and the panchromatic remote sensing image.

[0059] The multispectral remote sensing image is subjected to standardization processing, including:

[0060] Performing atmospheric correction on the multispectral remote sensing image can remove the influence of aerosol factors in the atmosphere on the reflectivity of ground objects.

[0061] The multispectral remote sensing image after atmospheric correction is resampled to unify the resolution of the multispectral remote sensing image after atmospheric correction to a preset value.

[0062] Resampling can ensure that the resolution of multispectral remote sensing images is consistent with other data (such as panchromatic images), ensuring data consistency, which is convenient for subsequent analysis and application. Commonly used resampling methods include nearest neighbor method, bilinear interpolation and bicubic interpolation.

[0063] The resampled multispectral remote sensing images are band fused, and the data of multiple channels are synthesized into the same multispectral remote sensing image to generate high-quality multispectral remote sensing images to provide support for subsequent analysis and application.

[0064] Common band fusion methods include simple channel splicing, principal component analysis (PCA), Brovey transform, etc.

[0065] The multispectral remote sensing image after band fusion is cut and divided into blocks according to the preset pixel size.

[0066] Define the size of each block, such as 128*128 pixels, and cut it according to the preset size. The remote sensing image after block division is helpful for the extraction and analysis of local features.

[0067] The panchromatic remote sensing image is subjected to standardization processing, including: cutting and dividing the panchromatic remote sensing image into blocks according to the preset pixel size.

[0068] Standardizing the multispectral remote sensing images and the panchromatic remote sensing images can ensure the consistency of remote sensing image data and provide a reliable data basis for subsequent analysis and application.

[0069] S103, inputting the standardized multispectral remote sensing image into a first semantic segmentation network, and the first semantic segmentation network performs semantic segmentation based on the standardized multispectral remote sensing image, and outputs the crop distribution result of the target farmer's land.

[0070] Among them, the first semantic segmentation network is obtained after training based on multispectral remote sensing image training samples.

[0071] The semantic segmentation network is a deep learning model that is used to classify each pixel in an image into a predefined category. Remote sensing semantic segmentation refers to the identification and classification of the ground area to be identified through shallow features such as spectrum, color, location information and abstract semantic deep features in remote sensing images, and giving it a real label. This requires both accurately depicting the outline of the segmented area and accurately determining the category corresponding to the area.

[0072] In remote sensing image processing, semantic segmentation is a method of analyzing image scenes, classifying and labeling each pixel in the image according to its semantic information, completing two tasks: pixel classification and image object segmentation. Common semantic segmentation networks include U-Net, DeepLab, and SegNet.

[0073] S104, inputting the standardized full-color remote sensing image into a second semantic segmentation network, and the second semantic segmentation network performs semantic segmentation according to the crop distribution result of the target farmer's land and the standardized full-color remote sensing image, and outputs the plot-level crop distribution result of the target farmer's land.

[0074] The second semantic segmentation network is obtained after training based on panchromatic remote sensing image training samples.

[0075] The structures of the first semantic segmentation network and the second semantic segmentation network both include:

[0076] An encoder, a decoder and a jump connection; a spatial attention mechanism is added to the encoder and the decoder respectively; the spatial attention mechanism highlights the local features of the space by dynamically adjusting the feature weights.

[0077] Exemplarily, this embodiment is based on the Pytorch deep learning framework, and adds a spatial attention mechanism on the basis of the U-Net semantic segmentation network. The encoder consists of a series of convolutional layers and pooling layers, which extract high-level semantic features while gradually reducing the spatial dimension of the input image. The decoder is symmetrical with the encoder and consists of a series of convolutional layers and upsampling layers. It is responsible for restoring the feature map extracted by the encoder to the semantic features of the same size as the input image. The jump connection mainly connects the features in the encoder with those in the decoder, which helps to restore the detailed information. The spatial attention mechanism fully exploits the regional information such as texture, context, spatial connection, etc. in the image features, and highlights the local features of the space and suppresses irrelevant features by dynamically adjusting the feature weights.

[0078] The representative model of the spatial attention mechanism is the Spatial Transformer Networks (STN). STN can transform various deformation data in space and automatically capture important regional features. It can ensure that after operations such as cropping, translation or rotation, the image can still obtain the same result as the original image before the operation. The STN network includes local networks, parameterized network sampling (network generator) and differential image sampling.

[0079] S105: Compare the plot-level crop distribution results of the target farmer's land with the target farmer's enclosure data to verify the target farmer's land assets.

[0080] If the crop types analyzed in the crop distribution results match the crop types provided by the enclosure data, and the ratio of the difference between the farmland area analyzed in the crop distribution results and the farmland area provided by the enclosure data to the farmland area provided by the enclosure data is less than a preset ratio, a first verification result for the target farmer is generated; if the crop types analyzed in the crop distribution results do not match the crop types provided by the enclosure data, or the ratio of the difference between the farmland area analyzed in the crop distribution results and the farmland area provided by the enclosure data to the farmland area provided by the enclosure data is greater than or equal to a preset ratio, a second verification result for the target farmer is generated.

[0081] Wherein, the first verification result indicates that the land assets of the target farmer have been automatically verified.

[0082] The second verification result indicates that a manual verification of the target farmer's land assets is required.

[0083] The present embodiment provides a farmer's land verification method based on multi-source remote sensing images, which obtains multi-source remote sensing images of target farmers, wherein the multi-source remote sensing images include multispectral remote sensing images and panchromatic remote sensing images. First, based on the multispectral remote sensing images, remote sensing semantic segmentation technology is used to identify regional crop planting conditions to obtain crop distribution results of the target farmer's land. In order to make up for the lack of details in the multispectral remote sensing images, semantic segmentation technology is used to extract rural land parcels based on the crop distribution results of the target farmer's land and the panchromatic remote sensing images to obtain plot-level crop distribution results of the target farmer's land. The plot-level crop distribution results of the target farmer's land are compared with the farmer's enclosure data to verify the authenticity of the farmer's land assets.

[0084] Compared with the farmer land verification method using a single remote sensing image, the semantic segmentation technology is used to improve the accuracy of the extracted farmer's farmland. The use of multispectral remote sensing images and panchromatic remote sensing images, in addition to being able to obtain the scope and area of ​​the farmer's land, can also obtain the farmer's crop information, so as to understand the farmer's information more comprehensively and accurately, and provide more effective help in verifying the authenticity of the farmer's land assets.

[0085] The first semantic segmentation network and the second semantic segmentation network in the farmer land verification method based on multi-source remote sensing images described in the above embodiments are obtained after training. The following embodiments provide a detailed introduction to the training process of the first semantic segmentation network and the second semantic segmentation network.

[0086] See also Figure 2 , which is a flow chart of another farmer's land verification method based on multi-source remote sensing images provided in an embodiment of the present application. Figure 2 As shown, the method comprises the following steps:

[0087] S201, obtaining multispectral remote sensing image training samples corresponding to the first semantic segmentation network and panchromatic remote sensing image training samples corresponding to the second semantic segmentation network.

[0088] The multispectral remote sensing image training samples include multispectral remote sensing image sample images and multispectral remote sensing image labels; the panchromatic remote sensing image training samples include panchromatic remote sensing images and panchromatic remote sensing image labels.

[0089] Multispectral remote sensing image sample maps refer to representative image fragments selected from the original multispectral remote sensing images. These sample maps contain information from multiple bands, each of which corresponds to a different electromagnetic wavelength range, thereby capturing different characteristics of the surface. For example, common multispectral images include four bands: red, green, blue, and near-infrared. These bands can be used to distinguish different ground objects such as vegetation, water bodies, and soil.

[0090] Multispectral remote sensing image labels are annotations of the types of objects in the sample images, which provide the "correct answers" required for supervised learning. Labels can be classification labels (for example, forests, farmlands, cities, etc.) or more complex structured labels (for example, pixel-level classification in semantic segmentation).

[0091] The panchromatic remote sensing image sample map is a single-band high-resolution image, which is mainly used to provide detailed surface information.

[0092] Panchromatic remote sensing image labels are annotations of the types of objects in the sample images, which provide the "correct answers" required for supervised learning. Labels can be classification labels, pixel-level labels, or bounding box labels.

[0093] Among the mainstream free and accessible multispectral satellite remote sensing data sources, the European Space Agency's Sentinel-2 satellite has become the main research data due to its advantages of full spectral band coverage, wide spatial resolution, short re-entry period, free access, and the presence of near-infrared bands that are sensitive to vegetation information. Its satellite image resolution is 10m, 20m, and 60m. For open source and free high-resolution remote sensing data, Google Earth data has become an important data source in the field of satellite remote sensing due to its global coverage, real-time updates, high resolution, and downloadability. Therefore, when collecting data, we mainly focus on Sentinel-2 satellite data and Google Earth data, and at the same time collect and organize crop characteristics and plot feature label data in the corresponding rural areas to facilitate subsequent semantic segmentation network training.

[0094] By preparing training samples, the performance of semantic segmentation networks based on multispectral remote sensing images and panchromatic remote sensing images can be significantly improved, enabling them to extract useful information from remote sensing images and perform tasks such as classification, segmentation, and target detection.

[0095] S202: performing standardization processing on the multispectral remote sensing image training samples and the panchromatic remote sensing image training samples.

[0096] Among them, the standardization processing of the multispectral remote sensing image training samples includes standardization processing of the multispectral remote sensing image sample map and standardization processing of the multispectral remote sensing image labels; the standardization processing of the multispectral remote sensing image labels includes resampling and geographic registration.

[0097] The standardization processing of the panchromatic remote sensing image training samples includes standardization processing of the panchromatic remote sensing image sample map and standardization processing of the panchromatic remote sensing image labels; the standardization processing of the panchromatic remote sensing image labels includes resampling and geographic registration.

[0098] After geographic registration, the coordinates of the remote sensing image labels and the remote sensing image sample maps are unified into the same coordinate system. Since the input data of the semantic segmentation network should not be too large, the processed remote sensing image labels and remote sensing image sample maps need to be cut into blocks, for example, cut into 256*256 pixel size, to form a training set.

[0099] S203. Based on the standardized multispectral remote sensing image training samples, determine the network parameters of the first semantic segmentation network through training and testing on the initial network structure of the first semantic segmentation network; based on the standardized panchromatic remote sensing image training samples, determine the network parameters of the second semantic segmentation network through training and testing on the initial network structure of the second semantic segmentation network.

[0100] The structures of the first semantic segmentation network and the second semantic segmentation network both include:

[0101] An encoder, a decoder and a jump connection; a spatial attention mechanism is added to the encoder and the decoder respectively; the spatial attention mechanism highlights the local features of the space by dynamically adjusting the feature weights.

[0102] The encoder consists of a series of convolutional layers and pooling layers, which extract high-level semantic features while gradually reducing the spatial dimension of the input image. The decoder is symmetrical to the encoder and consists of a series of convolutional layers and upsampling layers. It is responsible for restoring the feature maps extracted by the encoder to semantic features of the same size as the input image. The jump connection mainly connects the features in the encoder with those in the decoder, which helps to restore detailed information. The spatial attention mechanism fully mines the regional information such as texture, context, spatial connection, etc. in the image features, and highlights the local spatial features and suppresses irrelevant features by dynamically adjusting the feature weights.

[0103] For example, spatial transformer networks (STN) are added to the structures of the first semantic segmentation network and the second semantic segmentation network. STN can transform various deformation data in space and automatically capture important regional features. It can ensure that after the image is cropped, translated or rotated, the same result as the original image before the operation can be obtained.

[0104] The initial network structure of the first semantic segmentation network is trained using multispectral remote sensing image training samples, and the network parameters of the first semantic segmentation network are determined through hyperparameter tuning.

[0105] The initial network structure of the second semantic segmentation network is trained using panchromatic remote sensing image training samples, and the network parameters of the second semantic segmentation network are determined through hyperparameter tuning.

[0106] By adding a spatial attention mechanism to the semantic segmentation network and using training samples for training on this basis, the semantic segmentation network's ability to recognize features such as multi-scale, multi-dimensional spatial transformations and spatial contextual connections in remote sensing image data is enhanced.

[0107] S204. Acquire multi-source remote sensing images of the target farmer's land.

[0108] Acquire multi-source remote sensing images of the target farmer’s land, including multispectral remote sensing images and panchromatic remote sensing images of the area.

[0109] S205: performing standardization processing on the multispectral remote sensing image and the panchromatic remote sensing image.

[0110] The acquired multispectral remote sensing images and panchromatic remote sensing images are subjected to standardization processing such as atmospheric correction, resampling, and image segmentation to convert non-standard data into standard data for subsequent semantic segmentation network training.

[0111] S206, inputting the standardized multispectral remote sensing image into the trained first semantic segmentation network, and the trained first semantic segmentation network performs semantic segmentation based on the standardized multispectral remote sensing image, and outputs the crop distribution result of the target farmer's land.

[0112] The crop distribution results of the target farmers’ land reflect the planting distribution of the farmers’ land, including the types of crops planted and the scope of farmland.

[0113] S207, inputting the standardized full-color remote sensing image into a second semantic segmentation network, and the second semantic segmentation network performs semantic segmentation according to the crop distribution result of the target farmer's land and the standardized full-color remote sensing image, and outputs the plot-level crop distribution result of the target farmer's land.

[0114] Figure 3 The results of crop distribution on the land are shown. Figure 4 The plot-level crop distribution results of the land are shown. Compared with the crop distribution results of the target farmer's land, the plot-level crop distribution results of the target farmer's land reflect the plot distribution of the farmer's land in a more refined manner.

[0115] S208: Compare the plot-level crop distribution results of the target farmer's land with the target farmer's enclosure data to verify the target farmer's land assets.

[0116] Exemplarily, in this embodiment, farmers can enter information such as planting area, types of crops, etc. through relevant terminals, and define the scope of farmland by clicking on longitude and latitude coordinates. After confirmation, the information will be uploaded to the business system.

[0117] The business system will combine the longitude and latitude coordinates uploaded by farmers with the plot-level crop distribution results to identify the area of ​​farmland and the types of crops planted in the map.

[0118] The business system compares the farmland information obtained from the previous step with the information uploaded by the farmers. If the crop type in the uploaded information is correct and the difference in the planting area is less than 10%, the business system returns a verification result. If the above requirements are not met, the business personnel will be prompted that the authenticity of the farmland information is in doubt and will conduct manual verification.

[0119] For example, when banks carry out farmer loan business, they use the above method to identify the types and areas of crops planted by farmers, which allows them to have a more comprehensive understanding of farmers' information and improve the accuracy of the farmers' land assets. This, to a certain extent, solves the problem that farmers are too scattered for bank outlets to cover, improves convenience, and provides a basis for credit limit calculation and post-loan risk prevention and control.

[0120] Another farmer's land verification method based on multi-source remote sensing images provided in this embodiment shows that the use of remote sensing semantic segmentation technology to segment multi-source remote sensing images can improve the accuracy of extracting farmers' farmland. In addition to the scope and area of ​​farmers' farmland, it can also obtain farmers' crop information, provide a more comprehensive understanding of farmers' information, and provide more effective help in verifying the authenticity of farmers' land assets.

[0121] Compared with the classic semantic segmentation network, this application adds a spatial attention mechanism to the semantic segmentation network, which improves the semantic segmentation network's ability to recognize spatial transformation features and spatial context features in remote sensing images, and further improves the semantic segmentation network's segmentation capabilities for crop identification and fine-grained identification of plots.

[0122] Based on the farmer's land verification method based on multi-source remote sensing images introduced in the previous embodiment, the present application also provides a farmer's land verification device based on multi-source remote sensing images. Figure 5 Figure 1 is a schematic diagram of the structure of the device. Figure 5 As shown, the farmer land verification device based on multi-source remote sensing images includes:

[0123] A remote sensing image acquisition module 301 is used to acquire multi-source remote sensing images of target farmer's land;

[0124] The multi-source remote sensing images include multispectral remote sensing images and panchromatic remote sensing images.

[0125] A remote sensing image standardization processing module 302 is used to perform standardization processing on the multispectral remote sensing image and the panchromatic remote sensing image;

[0126] In an optional implementation, the multispectral remote sensing image is subjected to standardization processing, including:

[0127] Performing atmospheric correction on the multispectral remote sensing image;

[0128] Resampling the multispectral remote sensing image after atmospheric correction to unify the resolution of the multispectral remote sensing image after atmospheric correction to a preset value;

[0129] Perform band fusion on the resampled multispectral remote sensing images and synthesize the data of multiple channels into the same multispectral remote sensing image for use;

[0130] The multispectral remote sensing image after band fusion is cut and divided into blocks according to the preset pixel size.

[0131] In an optional implementation, the panchromatic remote sensing image is subjected to standardization processing, including:

[0132] The full-color remote sensing image is cut into blocks according to the preset pixel size.

[0133] The remote sensing image semantic segmentation module 303 is used to input the standardized multispectral remote sensing image into a first semantic segmentation network, and the first semantic segmentation network performs semantic segmentation based on the standardized multispectral remote sensing image, and outputs the crop distribution result of the target farmer's land; the first semantic segmentation network is obtained based on the multispectral remote sensing image training samples; the standardized panchromatic remote sensing image is input into a second semantic segmentation network, and the second semantic segmentation network performs semantic segmentation based on the crop distribution result of the target farmer's land and the standardized panchromatic remote sensing image, and outputs the plot-level crop distribution result of the target farmer's land; the second semantic segmentation network is obtained based on the panchromatic remote sensing image training samples;

[0134] The structures of the first semantic segmentation network and the second semantic segmentation network both include:

[0135] An encoder, a decoder and a jump connection; a spatial attention mechanism is added to the encoder and the decoder respectively; the spatial attention mechanism highlights the local features of the space by dynamically adjusting the feature weights.

[0136] Optionally, the farmer land verification device based on multi-source remote sensing images further includes a network training module, which is used to train the first semantic segmentation network in the following manner:

[0137] Acquire a multispectral remote sensing image training sample corresponding to the first semantic segmentation network; the multispectral remote sensing image training sample includes a multispectral remote sensing image sample map and a multispectral remote sensing image label;

[0138] The multispectral remote sensing image training samples are subjected to standardization processing; the standardization processing includes standardization processing of the multispectral remote sensing image sample map and standardization processing of the multispectral remote sensing image labels; the standardization processing of the multispectral remote sensing image labels includes resampling, geographic registration and cutting and blocking;

[0139] Based on the standardized multispectral remote sensing image training samples, the network parameters of the first semantic segmentation network are determined through training and testing on the initial network structure of the first semantic segmentation network.

[0140] Optionally, the farmer land verification device based on multi-source remote sensing images further includes a network training module, and the network training module is used to train the second semantic segmentation network in the following manner:

[0141] Obtaining a panchromatic remote sensing image training sample corresponding to the second semantic segmentation network; the panchromatic remote sensing image training sample includes a panchromatic remote sensing image and a panchromatic remote sensing image label;

[0142] The panchromatic remote sensing image training samples are standardized; the standardized processing includes standardized processing of the panchromatic remote sensing image sample map and standardized processing of the panchromatic remote sensing image labels; the standardized processing of the panchromatic remote sensing image labels includes resampling, geographic registration and cutting and blocking.

[0143] The farmer land verification module 304 is used to compare the plot-level crop distribution results of the target farmer's land with the enclosure data of the target farmer to verify the land assets of the target farmer.

[0144] If the crop types analyzed by the crop distribution results match the crop types provided by the enclosure data, and the difference between the farmland area analyzed by the crop distribution results and the farmland area provided by the enclosure data is less than a preset ratio, a first verification result of the target farmer is generated; if the crop types analyzed by the crop distribution results do not match the crop types provided by the enclosure data or the difference between the farmland area analyzed by the crop distribution results and the farmland area provided by the enclosure data is greater than or equal to a preset ratio, a second verification result of the target farmer is generated;

[0145] The first verification result indicates that the land assets of the target farmer have been automatically verified;

[0146] The second verification result indicates that a manual verification of the target farmer's land assets is required.

[0147] In addition, an embodiment of the present application also provides a computer-readable storage medium, in which a computer program is stored. When the program is executed by a processor, a farmer's land verification method based on multi-source remote sensing images as described in any of the method embodiments is implemented.

[0148] In addition, an embodiment of the present application also provides a processor for running a computer program, which, when running, executes the farmer land verification method based on multi-source remote sensing images as introduced in any implementation method of the aforementioned method embodiment.

[0149] It should be noted that each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device embodiment described above is merely schematic, in which the unit described as a separate component may or may not be physically separated, and the component prompted as a unit may or may not be a physical unit, that is, it may be located in one place, or it may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0150] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A farmer's land verification method based on multi-source remote sensing images, characterized in that: include: Acquire multi-source remote sensing images of the target farmer's land; the multi-source remote sensing images include multi-spectral remote sensing images and panchromatic remote sensing images; Performing standardization processing on the multispectral remote sensing image and the panchromatic remote sensing image; Inputting the standardized multispectral remote sensing image into a first semantic segmentation network, the first semantic segmentation network performs semantic segmentation based on the standardized multispectral remote sensing image, and outputs the crop distribution result of the target farmer's land; the first semantic segmentation network is obtained after training based on the multispectral remote sensing image training samples; The standardized panchromatic remote sensing image is input into a second semantic segmentation network, and the second semantic segmentation network performs semantic segmentation according to the crop distribution result of the target farmer's land and the standardized panchromatic remote sensing image, and outputs the plot-level crop distribution result of the target farmer's land; the second semantic segmentation network is obtained after training based on the panchromatic remote sensing image training samples; Compare the plot-level crop distribution results of the target farmer's land with the target farmer's enclosure data to verify the target farmer's land assets.

2. The method according to claim 1, characterized in that The structures of the first semantic segmentation network and the second semantic segmentation network both include: An encoder, a decoder and a jump connection; a spatial attention mechanism is added to the encoder and the decoder respectively; the spatial attention mechanism highlights the local features of the space by dynamically adjusting the feature weights.

3. The method according to claim 2, characterized in that The first semantic segmentation network is trained by the following steps: Acquire a multispectral remote sensing image training sample corresponding to the first semantic segmentation network; the multispectral remote sensing image training sample includes a multispectral remote sensing image sample map and a multispectral remote sensing image label; The multispectral remote sensing image training samples are subjected to standardization processing; the standardization processing includes standardization processing of the multispectral remote sensing image sample map and standardization processing of the multispectral remote sensing image labels; the standardization processing of the multispectral remote sensing image labels includes resampling, geographic registration and cutting and blocking; Based on the standardized multispectral remote sensing image training samples, the network parameters of the first semantic segmentation network are determined through training and testing on the initial network structure of the first semantic segmentation network.

4. The method according to claim 2, characterized in that: The second semantic segmentation network is trained by the following steps, including: Obtaining a panchromatic remote sensing image training sample corresponding to the second semantic segmentation network; the panchromatic remote sensing image training sample includes a panchromatic remote sensing image and a panchromatic remote sensing image label; The panchromatic remote sensing image training samples are subjected to standardization processing; the standardization processing includes standardization processing of the panchromatic remote sensing image sample map and standardization processing of the panchromatic remote sensing image labels; the standardization processing of the panchromatic remote sensing image labels includes resampling, geographic registration and cutting and blocking; Based on the standardized panchromatic remote sensing image training samples, the network parameters of the second semantic segmentation network are determined through training and testing on the initial network structure of the second semantic segmentation network.

5. The method according to claim 1, characterized in that: The multispectral remote sensing image is subjected to standardization processing, including: Performing atmospheric correction on the multispectral remote sensing image; Resampling the multispectral remote sensing image after atmospheric correction to unify the resolution of the multispectral remote sensing image after atmospheric correction to a preset value; Perform band fusion on the resampled multispectral remote sensing images and synthesize the data of multiple channels into the same multispectral remote sensing image for use; The multispectral remote sensing image after band fusion is cut and divided into blocks according to the preset pixel size.

6. The method according to claim 1, characterized in that The panchromatic remote sensing image is subjected to standardization processing, including: The panchromatic remote sensing image is cut and divided into blocks according to a preset pixel size.

7. The method according to claim 1, characterized in that The comparison of the plot-level crop distribution results of the target farmer's land and the enclosure data of the target farmer to verify the land assets of the target farmer includes: If the crop types analyzed by the crop distribution results match the crop types provided by the enclosure data, and the ratio of the difference between the farmland area analyzed by the crop distribution results and the farmland area provided by the enclosure data to the farmland area provided by the enclosure data is less than a preset ratio, a first verification result of the target farmer is generated; if the crop types analyzed by the crop distribution results do not match the crop types provided by the enclosure data, or the ratio of the difference between the farmland area analyzed by the crop distribution results and the farmland area provided by the enclosure data to the farmland area provided by the enclosure data is greater than or equal to a preset ratio, a second verification result of the target farmer is generated; The first verification result indicates that the land assets of the target farmer have been automatically verified; The second verification result indicates that a manual verification of the target farmer's land assets is required.

8. A farmer's land verification device based on multi-source remote sensing images, characterized in that: include: A remote sensing image acquisition module is used to acquire multi-source remote sensing images of the target farmer's land; the multi-source remote sensing images include multi-spectral remote sensing images and panchromatic remote sensing images; A remote sensing image standardization processing module, used for performing standardization processing on the multispectral remote sensing image and the panchromatic remote sensing image; A remote sensing image semantic segmentation module is used to input a standardized multispectral remote sensing image into a first semantic segmentation network, and the first semantic segmentation network performs semantic segmentation based on the standardized multispectral remote sensing image, and outputs a crop distribution result of a target farmer's land; the first semantic segmentation network is obtained after training based on multispectral remote sensing image training samples; the standardized panchromatic remote sensing image is input into a second semantic segmentation network, and the second semantic segmentation network performs semantic segmentation based on the crop distribution result of the target farmer's land and the standardized panchromatic remote sensing image, and outputs a plot-level crop distribution result of the target farmer's land; the second semantic segmentation network is obtained after training based on panchromatic remote sensing image training samples; The farmer land verification module is used to compare the plot-level crop distribution results of the target farmer's land with the target farmer's enclosure data to verify the target farmer's land assets.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the program is executed by the processor, the farmer's land verification method based on multi-source remote sensing images as described in any one of claims 1 to 7 is implemented.

10. A processor, characterized in that: Used to run a computer program, which, when running, executes the farmer's land verification method based on multi-source remote sensing images as described in any one of claims 1 to 7.

Images