Method, device and equipment for processing cross-regional crop remote sensing image fused with geographic information

Abstract

The application discloses a cross-region crop remote sensing image processing method, device and equipment fused with geographic information, relates to the technical field of crop data processing and remote sensing cross technology, constructs a generalization data set of latitude and terrain joint partitioning through sample geographic information data and sample satellite remote sensing image data, obtains a geographic prototype library corresponding to a geographic unit, then trains a crop classification prediction model based on the geographic prototype library, and performs crop classification prediction by using the crop classification prediction model, which can effectively capture the combined influence of geographic multidimensional factors on crop growth, solve the fundamental problem of poor generalization of the cross-region model, improve the cross-geographic processing capability of the crop classification prediction model, and realize high-precision, high-robustness and low-cost crop remote sensing classification.

Description

Technical Field

[0001] This application relates to the cross-disciplinary fields of crop data processing technology and remote sensing technology, specifically to a cross-regional crop remote sensing image processing method, apparatus, and equipment that integrates geographic information. Background Technology

[0002] With the increasing global demand for food security, remote sensing technology has become a core tool for crop classification and monitoring. Remote sensing data provides multi-dimensional observations of the Earth, including crucial spatial and temporal information, which is essential for addressing macro-monitoring tasks and, combined with artificial intelligence, enables accurate identification of large-scale crop types. Geographic information is critical in macro-monitoring tasks, such as crop classification. Crop growth characteristics and planting structures are significantly influenced by geographical environment (e.g., latitude, altitude, topography). Crop spectral characteristics and phenological stages are affected by differences in light intensity, accumulated temperature, and microclimate caused by altitude and slope, leading to a sharp decline in the generalization performance of models trained in a single region when applied across regions. Existing technologies do not fully quantify the coupling effects of geographical factors, making it difficult for models to adapt to complex geographical environments. Secondly, field sampling in high-altitude and remote areas is costly, and the significant differences in image data distribution across regions increase the difficulty of image-based crop information processing. Existing technologies typically rely on remote sensing data from a single geographical region to train crop classification models, depending on local sample labeling and failing to consider the impact of latitudinal differences on crop growth characteristics, making the models unsuitable for other regions. Summary of the Invention

[0003] The purpose of this application is to provide a cross-regional crop remote sensing image processing method, apparatus and equipment that integrates geographic information, which solves the problem that existing technologies usually rely on remote sensing data from a single geographic region to train crop classification models, depend on local sample annotation, do not consider the impact of latitude differences on crop growth characteristics, and the models cannot be applied to other regions.

[0004] This application is achieved through the following technical solution:

[0005] The first aspect of this application provides a cross-regional crop remote sensing image processing method that integrates geographic information, including:

[0006] The sample geographic information data and sample satellite remote sensing image data corresponding to the first target area are acquired, and the sample geographic information data and sample satellite remote sensing image data are preprocessed to obtain preprocessed data; wherein, the first target area represents the region where data needs to be collected.

[0007] Geographic units are divided and data are bound according to the preprocessed data to obtain multiple geographic units and crop sample data corresponding to each geographic unit. Based on the crop sample data corresponding to the geographic unit, a geographic prototype library corresponding to the geographic unit is established.

[0008] In response to a real-time crop prediction command input by a user through human-computer interaction, the target geographic unit corresponding to the second target area and its corresponding target geographic prototype library are determined; wherein, the second target area represents a sub-region determined in the first target area according to the real-time crop prediction command;

[0009] Based on the target geographic unit and its corresponding target geographic prototype library, a crop classification prediction model is trained, and real-time geographic information data and real-time satellite remote sensing image data corresponding to the second target area are collected. The crop classification prediction model is used to identify the real-time geographic information data and real-time satellite remote sensing image data to obtain crop classification prediction results.

[0010] In one possible implementation, after determining the target geographic unit corresponding to the second target area and its corresponding target geographic prototype library, the method further includes:

[0011] Determine the number of crop sample data in the target geographic prototype library, and determine the sufficiency of the samples based on the number of samples; the sufficiency of the samples includes whether the samples are sufficient or insufficient.

[0012] If the sample sufficiency is sufficient, no operation is performed, and the process proceeds to train the crop classification prediction model based on the target geographic unit and its corresponding target geographic prototype library.

[0013] If the sample sufficiency is insufficient, then based on the geographical phenological characteristics corresponding to the target geographical prototype library, similar geographical prototype libraries are matched in other geographical prototype libraries corresponding to the first target area and / or geographical prototype libraries corresponding to other areas; the other geographical prototype libraries refer to geographical prototype libraries that are different from the target geographical prototype library among all geographical prototype libraries corresponding to the first target area.

[0014] Based on the similar geographic prototype library, the target geographic prototype library is supplemented to obtain a supplemented target geographic prototype library. Then, based on the supplemented target geographic prototype library, the process of training a crop classification prediction model according to the target geographic unit and its corresponding target geographic prototype library is initiated.

[0015] In one possible implementation, acquiring sample geographic information data and sample satellite remote sensing image data corresponding to the first target area includes:

[0016] Obtain multispectral satellite imagery corresponding to the second target area to obtain sample satellite remote sensing image data;

[0017] Obtain the DEM data corresponding to the second target area to obtain sample geographic information data.

[0018] In one possible implementation, the sample geographic information data and sample satellite remote sensing image data are preprocessed to obtain preprocessed data, including:

[0019] The sample satellite remote sensing image data is preprocessed to obtain preprocessed sample satellite remote sensing image data; the image preprocessing includes radiometric calibration, orthorectification, image registration and / or image fusion.

[0020] The elevation data corresponding to the first target area is extracted from the sample geographic information data, and the slope data corresponding to the first target area is obtained based on the elevation characteristics.

[0021] Based on the elevation and slope data corresponding to the first target area, a geographic feature file divided by cultivated land boundaries is used for partitioning calculation to obtain the elevation and slope data corresponding to each crop sample area in the first target area.

[0022] The elevation data and slope data are written into the attribute table of the crop sample area in the geographic feature file to obtain the geographic feature file with the crop sample area attributes written in, thus obtaining the first target geographic feature file.

[0023] Based on the smallest outer square of each crop sample area, the preprocessed sample satellite remote sensing image data is segmented to obtain the slice multispectral image corresponding to each crop sample area;

[0024] The first target geographic feature file and the sliced ​​multispectral image corresponding to each crop sample area are used together as preprocessed data.

[0025] In one possible implementation, the preprocessed data is used to divide geographic units and bind data to obtain multiple geographic units and crop sample data corresponding to each geographic unit, including:

[0026] Based on the spatial coordinate system corresponding to the preprocessed sample satellite remote sensing image data, the latitude range corresponding to the first target area is extracted;

[0027] Based on the latitude range corresponding to the first target region, the first target region is divided into horizontal intervals using unit latitude as the basis, so as to divide the first target region into multiple latitude zones and obtain a two-dimensional network; wherein, the unit latitude is set to 1°.

[0028] Based on the altitude range corresponding to the first target area, the first target area is vertically divided into multiple independent geographical units, with each unit altitude as the basis; wherein, the unit altitude is set to 200m.

[0029] Based on the coordinates corresponding to the geographic unit, the latitude range and altitude range corresponding to the geographic unit are written into the first target geographic element file to obtain the second target geographic element file; wherein, the first target geographic element file uses the same spatial coordinate system as the preprocessed sample satellite remote sensing image data;

[0030] For any given geographic unit, the target crop sample area located within the latitude and altitude range corresponding to the geographic unit is determined based on the second target geographic feature file. The crop sample area is then constructed by combining the tiled multispectral image, latitude data, altitude data, slope data, data sampling time, and corresponding crop label data. This yields the crop sample data corresponding to the geographic unit. The crop label data consists of real crop labels input through human-computer interaction.

[0031] Traverse all geographic units to obtain crop sample data corresponding to each geographic unit.

[0032] In one possible implementation, after obtaining multiple geographic units and crop sample data corresponding to each geographic unit, the method further includes: performing a time consistency check on the crop sample data, identifying crop sample data that failed the check, and marking the crop sample data that failed the check.

[0033] In one possible implementation, a crop classification prediction model is trained based on the target geographic unit and its corresponding target geographic prototype library, including:

[0034] Based on the target crop sample data in the target geographic prototype library corresponding to the target geographic unit, extract the latitude data, altitude data, slope data, and the latitude vector, altitude vector, slope vector, and time vector corresponding to the data sampling time from the target crop sample data.

[0035] The multispectral images of the target crop sample data are stitched together with latitude vector, altitude vector, slope vector and time vector to obtain sample fusion feature data;

[0036] Using the sample fusion feature data as the actual input and the crop label data in the target crop sample data as the expected output, a crop classification prediction model is trained.

[0037] In one possible implementation, real-time geographic information data and real-time satellite remote sensing image data corresponding to the second target area are collected. A crop classification prediction model is then used to identify the real-time geographic information data and real-time satellite remote sensing image data to obtain crop classification prediction results, including:

[0038] Collect real-time geographic information data and real-time satellite remote sensing image data corresponding to the second target area;

[0039] Based on the real-time geographic information data and real-time satellite remote sensing image data, determine the real-time slice multispectral image, real-time latitude data, real-time altitude data, real-time slope data, and real-time data sampling time corresponding to the second target area;

[0040] Based on the real-time sliced ​​multispectral image, real-time latitude data, real-time altitude data, real-time slope data, and real-time data sampling time, real-time fused feature data is obtained;

[0041] The crop classification prediction model is used to identify the real-time fused feature data to obtain crop classification prediction results.

[0042] Based on the same inventive concept, a second aspect of this application provides a cross-regional crop remote sensing image processing device that integrates geographic information, comprising:

[0043] The sample data preprocessing module is used to acquire sample geographic information data and sample satellite remote sensing image data corresponding to the first target area, and to preprocess the sample geographic information data and sample satellite remote sensing image data to obtain preprocessed data; wherein, the first target area represents the region where data needs to be collected.

[0044] The prototype library construction module is used to divide geographical units and bind data according to the preprocessed data to obtain multiple geographical units and crop sample data corresponding to each geographical unit, and to build a geographical prototype library corresponding to the geographical unit based on the crop sample data corresponding to the geographical unit.

[0045] The prediction instruction response module is used to respond to a real-time crop prediction instruction input by the user through human-computer interaction, and to determine the target geographic unit corresponding to the second target area and its corresponding target geographic prototype library; wherein, the second target area represents the area determined in the first target area according to the real-time crop prediction instruction;

[0046] The crop classification prediction module is used to train a crop classification prediction model based on the target geographic unit and its corresponding target geographic prototype library, and to collect real-time geographic information data and real-time satellite remote sensing image data corresponding to the second target area. The crop classification prediction model is then used to identify the real-time geographic information data and real-time satellite remote sensing image data to obtain crop classification prediction results.

[0047] Based on the same inventive concept, a third aspect of this application provides an electronic device, including a memory and a processor, wherein the memory is used to store a set of computer instructions; when the processor executes the set of computer instructions, it performs the operation steps of the method described in any possible embodiment of the first aspect above.

[0048] Compared with the prior art, this application has the following advantages and beneficial effects:

[0049] This application provides a cross-regional crop remote sensing image processing method, apparatus, and device that integrates geographic information. It constructs a generalized dataset with joint latitude and topography partitioning using sample geographic information data and sample satellite remote sensing image data, obtaining a geographic prototype library corresponding to each geographic unit. Then, a crop classification prediction model is trained based on this geographic prototype library, and crop classification prediction is performed using the model. This effectively captures the combined influence of multidimensional geographic factors on crop growth, solves the fundamental problem of poor generalization in cross-regional models, improves the cross-geographic processing capability of crop classification prediction models, and achieves high-precision, high-robustness, and low-cost crop remote sensing classification. Attached Figure Description

[0050] To more clearly illustrate the technical solutions of the exemplary embodiments of this application, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this application and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0051] Figure 1 A flowchart illustrating a cross-regional crop remote sensing image processing method that integrates geographic information, provided as an embodiment of this application;

[0052] Figure 2 A schematic diagram of the structure of a cross-regional crop remote sensing image processing device that integrates geographic information, provided in an embodiment of this application;

[0053] Figure 3 A schematic diagram of the structure of a cross-regional crop remote sensing image processing device that integrates geographic information, provided in an embodiment of this application;

[0054] The attached diagram shows the markings and corresponding component names:

[0055] 201-Sample data preprocessing module, 202-Prototype library construction module, 203-Prediction command response module, 204-Crop classification prediction module, 301-Memory, 302-Processor, 303-Bus. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the embodiments and accompanying drawings. The illustrative embodiments and descriptions of this application are only for explaining this application and are not intended to limit this application.

[0057] like Figure 1 As shown in the figure, this application provides a cross-regional crop remote sensing image processing method that integrates geographic information, including:

[0058] S101. Obtain sample geographic information data and sample satellite remote sensing image data corresponding to the first target area, and preprocess the sample geographic information data and sample satellite remote sensing image data to obtain preprocessed data; wherein, the first target area represents the region where data needs to be collected.

[0059] The sample geographic information data corresponding to the first target area can refer to terrain-related data, such as DEM (Digital Elevation Model) data. A DEM is a digital model that uses numerical arrays to represent surface elevation information. Through regular grids (such as raster grids) or irregular triangular meshes (TINs), the undulations of the Earth's surface are abstracted into a quantifiable dataset, with each pixel or node corresponding to a specific geographic coordinate elevation value. The core value of a DEM lies in its three-dimensional digital representation of terrain, which can not only intuitively present landforms such as mountains, valleys, and plains, but also extract parameters such as slope, aspect, and topographic relief through derivative analysis. Its data sources are diverse, including satellite remote sensing, aerial photogrammetry, lidar scanning, and ground measurements, with accuracy ranging from meters to centimeters. As fundamental data for geographic information systems, DEMs are widely used in fields such as geological hazard assessment (e.g., landslide simulation), hydrological analysis (watershed delineation), urban planning (earthwork calculation), agricultural layout, and military navigation. For example, in flood warning, DEMs can accurately simulate the inundation area; in transportation planning, they can optimize route selection to reduce engineering costs.

[0060] Therefore, the sample geographic information data and sample satellite remote sensing image data can be preprocessed to extract the terrain features and graphic features corresponding to the crop sample area, thereby obtaining preprocessed data. The crop sample area can refer to a complete arable land in the first target area.

[0061] S102. Based on the preprocessed data, the geographic units are divided and data are bound to obtain multiple geographic units and crop sample data corresponding to each geographic unit. Based on the crop sample data corresponding to the geographic unit, a geographic prototype library corresponding to the geographic unit is established.

[0062] By dividing geographic units and binding data, it is possible not only to predict crop classification in any sub-region of the first target area, but also to supplement samples based on geographic information when crop sample data in any sub-region of the first target area is insufficient.

[0063] For each geographic unit, the typical feature library contains crop sample data corresponding to all crop sample areas of the geographic unit. Therefore, when performing crop classification prediction on any sub-region within the first target region, a geographic prototype library corresponding to any sub-region can be determined, and training can be performed using this geographic prototype library. This allows for the effective learning of the terrain and graphic features of any sub-region, thereby achieving more accurate crop classification prediction.

[0064] When determining the geographic unit to which crop sample data belongs, there may be a crop sample area corresponding to one crop sample data that is located in two or more geographic units at the same time. In this case, we can determine which geographic unit the crop sample area corresponding to the crop sample data has the largest proportion, and then determine which geographic unit the crop sample data belongs to.

[0065] S103. In response to a real-time crop prediction command input by the user through human-computer interaction, determine the target geographic unit corresponding to the second target area and its corresponding target geographic prototype library; wherein, the second target area represents a sub-region determined in the first target area according to the real-time crop prediction command;

[0066] Real-time crop forecasting instructions can include instructions to determine a second target region within a first target region and to perform crop classification predictions within the second target region. The second target region can be part or all of the first target region. By determining the target geographic units corresponding to the second target region and their corresponding target geographic prototype libraries, basic data for crop classification can be obtained.

[0067] It is worth noting that determining the second target region within the first target region for crop classification prediction is merely a preferred embodiment of this application. Alternatively, all geographic prototype libraries of the first target region can be directly determined, and training and prediction can be performed based on these libraries, which can also achieve crop classification prediction for the second target region.

[0068] S104. Based on the target geographic unit and its corresponding target geographic prototype library, train a crop classification prediction model, and collect real-time geographic information data and real-time satellite remote sensing image data corresponding to the second target area. Use the crop classification prediction model to identify the real-time geographic information data and real-time satellite remote sensing image data to obtain crop classification prediction results.

[0069] After training the crop classification prediction model, the model has the ability to identify crop sample data. Therefore, it can construct real-time data with the same structure as crop sample data by using real-time geographic information data and real-time satellite remote sensing image data corresponding to the second target area. Then, the crop classification prediction model can identify the real-time data to achieve crop classification prediction.

[0070] This application provides a cross-regional crop remote sensing image processing method, apparatus, and device that integrates geographic information. It constructs a generalized dataset with joint latitude and topography partitioning using sample geographic information data and sample satellite remote sensing image data, obtaining a geographic prototype library corresponding to each geographic unit. Then, a crop classification prediction model is trained based on this geographic prototype library, and crop classification prediction is performed using the model. This effectively captures the combined influence of multidimensional geographic factors on crop growth, solves the fundamental problem of poor generalization in cross-regional models, improves the cross-geographic processing capability of crop classification prediction models, and achieves high-precision, high-robustness, and low-cost crop remote sensing classification.

[0071] In one possible implementation, after determining the target geographic unit corresponding to the second target area and its corresponding target geographic prototype library, the method further includes:

[0072] Determine the number of crop sample data in the target geographic prototype library, and determine the sufficiency of the samples based on the number of samples; the sufficiency of the samples includes whether the samples are sufficient or insufficient.

[0073] For example, a sample size threshold can be preset. If the number of crop sample data in the target geographic prototype library is less than the sample size threshold, the sample sufficiency is determined to be insufficient; otherwise, the sample sufficiency is determined to be sufficient.

[0074] If the sample sufficiency is sufficient, no operation is performed, and the process proceeds to train the crop classification prediction model based on the target geographic unit and its corresponding target geographic prototype library.

[0075] If the sample sufficiency is sufficient, it indicates that the crop classification prediction model trained based on the target geographic prototype library can accurately predict crop classification, and the process of training the crop classification prediction model based on the target geographic unit and its corresponding target geographic prototype library can be directly entered.

[0076] If the sample sufficiency is insufficient, then based on the geographical phenological characteristics corresponding to the target geographical prototype library, similar geographical prototype libraries are matched in other geographical prototype libraries corresponding to the first target area and / or geographical prototype libraries corresponding to other areas; the other geographical prototype libraries refer to geographical prototype libraries that are different from the target geographical prototype library among all geographical prototype libraries corresponding to the first target area.

[0077] Based on the similar geographic prototype library, the target geographic prototype library is supplemented to obtain a supplemented target geographic prototype library. Then, based on the supplemented target geographic prototype library, the process of training a crop classification prediction model according to the target geographic unit and its corresponding target geographic prototype library is initiated.

[0078] For example, a geographic prototype library can be established first: due to latitudinal differences, the gradient of heat conditions varies significantly, leading to crop distribution patterns and phenological changes that vary with latitude. In addition, the increase in altitude and topographic changes also have a significant impact on crop distribution and phenological patterns; phenological periods are delayed with increasing altitude, forming three-dimensional phenology. This application's embodiment stores typical characteristic prototypes of each latitude-geographic unit through the constructed geographic prototype library. , These represent the phenological sample data corresponding to the 1st, 2nd, ..., nth crop sample data in the geographic unit, where n represents the total number of phenological sample data in the geographic unit. Each phenological sample data contains geographic phenological information and crop category label.

[0079] Then, neighboring sample supplementation can be performed: when the sample sufficiency is insufficient, geographic information-driven compensation is used. First, some crop prototype samples are classified based on prior three-dimensional phenological patterns. Then, nearest neighbor retrieval (K-NN) is used to match similar prototypes to generate supplementary samples and labels. When a new regional task with few or no samples appears, the required crop sample data is supplemented in similar geographic units obtained from nearest neighbor retrieval. The specific method is as follows:

[0080] Constructing geographic phenological feature vectors from phenological sample data The geographic phenological feature vector adopts a 4-dimensional structure, denoted as... =[ , , ] T , It refers to altitude. It's longitude. It's latitude. It represents the shooting time, and T stands for transpose.

[0081] To eliminate the interference of differences in feature dimensions on similarity calculation, the geographic phenological feature values ​​in the above geographic phenological feature vector need to be standardized and encoded. The Z-score standardization method is used for standardization, and the formula is as follows:

[0082]

[0083] in, It is the i-th original geographical phenological feature value in the geographical phenological feature vector, i=1,2,3,4; These are standardized geographical phenological characteristic values. The mean of the i-th original geographical phenological feature values ​​corresponding to all phenological sample data in the typical feature prototype is given. is the standard deviation of the i-th original geographic phenological feature value corresponding to all phenological sample data in the typical feature prototype;

[0084] Based on the standardized geographical phenological feature values, the average feature value corresponding to the target geographical unit is obtained, forming a query vector that can represent the overall geographical environment of the region, and a feature vector that represents the geographical phenological attributes of the target region.

[0085]

[0086] in, It is a query vector. It is the average value of the first characteristic. It is the average value of the second characteristic. It is the average value of the third characteristic. It is the average value of the fourth characteristic.

[0087] For the target geographic unit, the search space can be limited to a similar geographic prototype library corresponding to the target crop category; wherein, the similar geographic prototype library is different from the target geographic prototype library, and the similar geographic prototype library includes similar crop sample data and corresponding similar typical feature prototypes, denoted as . ;

[0088] For example, you can first match all geographic units similar to the target geographic unit by crop type; then collect all crop samples under these similar geographic units; next, calculate the Euclidean distance between the query vector and these samples; finally, select the K samples with the smallest distance as supplementary samples.

[0089] For any one Get the query vector and The Euclidean distance between the geographic phenological feature vectors corresponding to each phenological sample data is:

[0090]

[0091] in, It is the Euclidean distance between the query vector and the geographic phenological feature vector corresponding to each phenological sample data in the similar typical feature prototype. It is the i-th feature value in the query vector. It is the i-th feature value of the phenological sample data in the prototype of similar typical features. It is an optional feature weighting coefficient used to adjust the importance of different geographical phenological feature values ​​in similarity measurement.

[0092] Select the K phenological sample data with the smallest Euclidean distance from the similar typical feature prototypes, and construct the selected phenological sample data. We obtain a set of similar samples, denoted as . This can be expressed as a formula: ;in, These represent the 1st, 2nd, ..., Kth selected phenological sample data, where K represents a pre-set positive integer.

[0093] Based on the set of similar samples, the corresponding similar crop sample data are used as new training sample features and added to the target geographic prototype library corresponding to the second target area to complete the sample supplementation.

[0094] In one possible implementation, acquiring sample geographic information data and sample satellite remote sensing image data corresponding to the first target area includes:

[0095] Multispectral satellite imagery corresponding to the second target area is acquired to obtain sample satellite remote sensing image data.

[0096] Obtain the DEM data corresponding to the second target area to obtain sample geographic information data.

[0097] In one possible implementation, the sample geographic information data and sample satellite remote sensing image data are preprocessed to obtain preprocessed data, including:

[0098] The sample satellite remote sensing image data is preprocessed to obtain preprocessed sample satellite remote sensing image data; the image preprocessing includes radiometric calibration, orthorectification, image registration and / or image fusion.

[0099] It is worth noting that the image preprocessing method described in the embodiments of this application is merely an example. Other image preprocessing methods may also be used based on the technical solutions provided in the embodiments of this application.

[0100] The elevation data corresponding to the first target area is extracted from the sample geographic information data, and the slope data corresponding to the first target area is obtained based on the elevation characteristics.

[0101] Based on the elevation and slope data corresponding to the first target area, a geographic feature file divided by cultivated land boundaries is used for partitioning calculation to obtain the elevation and slope data corresponding to each crop sample area in the first target area.

[0102] Among them, the geographic feature file delineated by cultivated land boundaries is a file stored in shp format, which can realize the rapid determination and division of crop sample areas.

[0103] The elevation and slope data are written into the attribute table of the crop sample area in the geographic feature file to obtain the geographic feature file containing the attributes of the crop sample area, thus obtaining the first target geographic feature file.

[0104] Based on the smallest outer square of each crop sample area, the preprocessed sample satellite remote sensing image data is segmented to obtain the sliced ​​multispectral image corresponding to each crop sample area.

[0105] The first target geographic feature file and the sliced ​​multispectral image corresponding to each crop sample area are used together as preprocessed data.

[0106] The preprocessed sample satellite remote sensing image data and the first target geographic feature file have the same spatial coordinate system, which facilitates the segmentation of geographic units.

[0107] In one possible implementation, the preprocessed data is used to divide geographic units and bind data to obtain multiple geographic units and crop sample data corresponding to each geographic unit, including:

[0108] Based on the spatial coordinate system corresponding to the preprocessed sample satellite remote sensing image data, the latitude range corresponding to the first target area is extracted.

[0109] Based on the spatial coordinate system corresponding to the preprocessed sample satellite remote sensing image data, the latitude range corresponding to the first target area is extracted according to the northernmost and southernmost effective pixel coordinates of the image, and the altitude range corresponding to the first target area is extracted according to the maximum and minimum values ​​of the altitude data.

[0110] Based on the latitude range corresponding to the first target region, the first target region is divided into horizontal intervals using unit latitude as the basis, thereby dividing the first target region into multiple latitudinal zones to obtain a two-dimensional network. The unit latitude is set to 1°.

[0111] Based on the altitude range corresponding to the first target area, the first target area is vertically divided into multiple independent geographical units, using a unit altitude as the basis. The unit altitude is set to 200m.

[0112] After dividing the geographic units, the latitude and altitude ranges corresponding to each geographic unit can be determined. Then, based on the latitude and altitude ranges corresponding to the geographic units, it can be determined which geographic unit the crop sample area is located in.

[0113] Based on the coordinates corresponding to the geographic unit, the latitude and altitude ranges corresponding to the geographic unit are written into a first target geographic element file to obtain a second target geographic element file. The first target geographic element file uses the same spatial coordinate system as the preprocessed sample satellite remote sensing image data.

[0114] For any given geographic unit, the target crop sample area is determined based on the second target geographic feature file, within the latitude and altitude range corresponding to the geographic unit. The crop sample area is then constructed by combining the corresponding multispectral imagery, latitude data, altitude data, slope data, data sampling time, and crop label data, resulting in the crop sample data for the geographic unit. The crop label data consists of real crop labels input through human-computer interaction.

[0115] Traverse all geographic units to obtain crop sample data corresponding to each geographic unit.

[0116] Optionally, the crop sample data corresponding to the geographic unit is structured data, so it can be stored in JSON format.

[0117] In one possible implementation, after obtaining multiple geographic units and crop sample data corresponding to each geographic unit, the method further includes: performing a time consistency check on the crop sample data, identifying crop sample data that failed the check, and marking the crop sample data that failed the check.

[0118] The image capture time is extracted from the metadata of the multispectral image, and the growth period of the corresponding plot is extracted from the crop label data table. The image capture time must fall within the crop growth period to ensure sample validity and time label consistency. Labeling crop sample data that fails validation avoids using incorrect crop sample data for model training, effectively ensuring the accuracy of the final crop classification prediction results.

[0119] In one possible implementation, a crop classification prediction model is trained based on the target geographic unit and its corresponding target geographic prototype library, including:

[0120] Based on the target crop sample data in the target geographic prototype library corresponding to the target geographic unit, extract the latitude data, altitude data, slope data, and the latitude vector, altitude vector, slope vector, and time vector corresponding to the data sampling time from the target crop sample data.

[0121] The multispectral images of the target crop sample data are stitched together with latitude vector, altitude vector, slope vector and time vector to obtain sample fusion feature data.

[0122] Using the sample fusion feature data as the actual input and the crop label data in the target crop sample data as the expected output, a crop classification prediction model is trained.

[0123] For example, during the geographic unit segmentation stage, the multidimensional geographic information has been discretized and binned for indexing. Latitude is binned at 1 degree (latitude range: [-90, 90]), altitude at 200m (altitude range: [0, 4000]), slope at 5 degrees (slope range: [0, 90]), and time at 1 month. After indexing, each bin is processed through an embedding layer to obtain corresponding vector representations: LatEmb represents the latitude vector after embedding, HeightEmb represents the altitude vector after embedding, SlopeEmb represents the slope vector after embedding, and TimeEmb represents the time vector after embedding. All these vectors have the same dimension, D. Concatenating the embedding vectors of the geographic information yields a geographic information vector with dimension 4*D.

[0124]

[0125] in, Represents a geographic information vector;

[0126] The geographic information vector serves as additional information for each input, and is concatenated with the image features corresponding to the tiled multispectral image (dimg in dimension; these features can be obtained by first dividing the image into fixed-size blocks, converting the image blocks into embedding vectors, then superimposing location encodings, and finally processing them using a self-attention mechanism, or by using a deep learning model to extract image features). The concatenation operation adds these features to the last dimension of the feature vector, resulting in a fused feature vector of dimension Dimg + 4*D.

[0127] FusedFeat = Concat(ImgFeat, GeoEmb)

[0128] Where FusedFeat represents the fused feature vector and ImgFeat represents the image feature.

[0129] The fused feature vectors are linearly projected onto the Transformer encoder for training. During training, geographic information encoding and terrain-driven compensation are used to improve the Transformer encoder's cross-geographic processing capabilities, and the Adam optimizer is employed for training to obtain a crop classification prediction model.

[0130] In one possible implementation, real-time geographic information data and real-time satellite remote sensing image data corresponding to the second target area are collected. A crop classification prediction model is then used to identify the real-time geographic information data and real-time satellite remote sensing image data to obtain crop classification prediction results, including:

[0131] Collect real-time geographic information data and real-time satellite remote sensing image data corresponding to the second target area;

[0132] Based on the real-time geographic information data and real-time satellite remote sensing image data, determine the real-time slice multispectral image, real-time latitude data, real-time altitude data, real-time slope data, and real-time data sampling time corresponding to the second target area;

[0133] Based on the real-time sliced ​​multispectral image, real-time latitude data, real-time altitude data, real-time slope data, and real-time data sampling time, real-time fused feature data is obtained;

[0134] The crop classification prediction model is used to identify the real-time fused feature data to obtain crop classification prediction results.

[0135] The crop classification prediction model outputs crop classification results, which is the probability distribution of crops in the crop sample area belonging to different crop types. By performing the .argmax() operation on the probability distribution, the predicted crop type label can be obtained, realizing the classification prediction of crop type, and adding the classification sample with the highest confidence to the geographic prototype library.

[0136] like Figure 2 As shown, based on the same inventive concept, embodiments of this application provide a cross-regional crop remote sensing image processing device that integrates geographic information, comprising:

[0137] The sample data preprocessing module 201 is used to acquire sample geographic information data and sample satellite remote sensing image data corresponding to the first target area, and to preprocess the sample geographic information data and sample satellite remote sensing image data to obtain preprocessed data; wherein, the first target area represents the region where data needs to be collected.

[0138] The prototype library construction module 202 is used to divide geographical units and bind data according to the preprocessed data to obtain multiple geographical units and crop sample data corresponding to each geographical unit, and to establish a geographical prototype library corresponding to the geographical unit based on the crop sample data corresponding to the geographical unit.

[0139] The prediction instruction response module 203 is used to respond to a real-time crop prediction instruction input by a user through human-computer interaction, and to determine the target geographic unit corresponding to the second target area and its corresponding target geographic prototype library; wherein, the second target area represents the area determined in the first target area according to the real-time crop prediction instruction;

[0140] The crop classification prediction module 204 is used to train a crop classification prediction model based on the target geographic unit and its corresponding target geographic prototype library, and to collect real-time geographic information data and real-time satellite remote sensing image data corresponding to the second target area. The crop classification prediction model is used to identify the real-time geographic information data and real-time satellite remote sensing image data to obtain crop classification prediction results.

[0141] The cross-regional crop remote sensing image processing device that integrates geographic information provided in this application embodiment can execute the above-mentioned method and technical solution. Its principle and beneficial effects are similar, and will not be described again here.

[0142] like Figure 3 As shown, based on the same inventive concept, this application provides an electronic device that may include a memory 301 and a processor 302. Exemplarily, the memory 301 and the processor 302 are interconnected via a bus 303.

[0143] The memory 301 stores computer-executed instructions;

[0144] The processor 302 executes the computer execution instructions stored in the memory, causing the processor to perform any of the cross-regional crop remote sensing image processing methods that integrate geographic information as described above.

[0145] This invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement any of the above-described cross-regional crop remote sensing image processing methods that integrate geographic information.

[0146] This invention can also provide a computer program product, including a computer program that, when executed by a processor, implements any of the above-described cross-regional crop remote sensing image processing methods that integrate geographic information.

[0147] All or part of the steps in the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a readable memory. When the program is executed, it performs the steps of the above method embodiments; and the aforementioned memory (storage medium) includes: read-only memory (ROM), RAM, flash memory, hard disk, solid-state drive, magnetic tape, floppy disk, optical disk, and any combination thereof.

[0148] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processing unit of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processing unit of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0149] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0150] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0151] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the claims.

[0152] It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A cross-regional crop remote sensing image processing method integrating geographic information, characterized in that, include: The sample geographic information data and sample satellite remote sensing image data corresponding to the first target area are acquired, and the sample geographic information data and sample satellite remote sensing image data are preprocessed to obtain preprocessed data; wherein, the first target area represents the region where data needs to be collected; Geographic units are divided and data are bound according to the preprocessed data to obtain multiple geographic units and crop sample data corresponding to each geographic unit. Based on the crop sample data corresponding to the geographic unit, a geographic prototype library corresponding to the geographic unit is established. In response to a real-time crop prediction command input by a user through human-computer interaction, the target geographic unit corresponding to the second target area and its corresponding target geographic prototype library are determined; wherein, the second target area represents a sub-region determined in the first target area according to the real-time crop prediction command; Based on the target geographic unit and its corresponding target geographic prototype library, a crop classification prediction model is trained, and real-time geographic information data and real-time satellite remote sensing image data corresponding to the second target area are collected. The crop classification prediction model is used to identify the real-time geographic information data and real-time satellite remote sensing image data to obtain crop classification prediction results. After determining the target geographic units corresponding to the second target area and their corresponding target geographic prototype library, the following steps are also included: Determine the number of crop sample data in the target geographic prototype library, and determine the sufficiency of the samples based on the number of samples; the sufficiency of the samples includes whether the samples are sufficient or insufficient. If the sample sufficiency is sufficient, no operation is performed, and the process proceeds to train the crop classification prediction model based on the target geographic unit and its corresponding target geographic prototype library. If the sample sufficiency is insufficient, then based on the geographical phenological characteristics corresponding to the target geographical prototype library, similar geographical prototype libraries are matched in other geographical prototype libraries corresponding to the first target area and / or geographical prototype libraries corresponding to other areas; the other geographical prototype libraries refer to geographical prototype libraries that are different from the target geographical prototype library among all geographical prototype libraries corresponding to the first target area. Based on the similar geographic prototype library, the target geographic prototype library is supplemented to obtain a supplemented target geographic prototype library. Then, based on the supplemented target geographic prototype library, the process of training a crop classification prediction model according to the target geographic unit and its corresponding target geographic prototype library begins.

2. The cross-regional crop remote sensing image processing method integrating geographic information according to claim 1, characterized in that, Acquire sample geographic information data and sample satellite remote sensing image data corresponding to the first target area, including: Obtain multispectral satellite imagery corresponding to the second target area to obtain sample satellite remote sensing image data; Obtain the DEM data corresponding to the second target area to obtain sample geographic information data.

3. The cross-regional crop remote sensing image processing method integrating geographic information according to claim 2, characterized in that, The sample geographic information data and sample satellite remote sensing image data are preprocessed to obtain preprocessed data, including: The sample satellite remote sensing image data is preprocessed to obtain preprocessed sample satellite remote sensing image data; the image preprocessing includes radiometric calibration, orthorectification, image registration and / or image fusion. The elevation data corresponding to the first target area is extracted from the sample geographic information data, and the slope data corresponding to the first target area is obtained based on the elevation data. Based on the elevation and slope data corresponding to the first target area, a geographic feature file divided by cultivated land boundaries is used for partitioning calculation to obtain the elevation and slope data corresponding to each crop sample area in the first target area. The elevation data and slope data are written into the attribute table of the crop sample area in the geographic feature file to obtain the geographic feature file with the crop sample area attributes written in, thus obtaining the first target geographic feature file. Based on the smallest outer square of each crop sample area, the preprocessed sample satellite remote sensing image data is segmented to obtain the slice multispectral image corresponding to each crop sample area; The first target geographic feature file and the sliced ​​multispectral image corresponding to each crop sample area are used together as preprocessed data.

4. The cross-regional crop remote sensing image processing method integrating geographic information according to claim 3, characterized in that, Based on the preprocessed data, geographical units are divided and data binding is performed to obtain multiple geographical units and crop sample data corresponding to each geographical unit, including: Based on the spatial coordinate system corresponding to the preprocessed sample satellite remote sensing image data, the latitude range corresponding to the first target area is extracted; Based on the latitude range corresponding to the first target region, the first target region is divided into horizontal intervals using unit latitude as the basis, so as to divide the first target region into multiple latitude zones and obtain a two-dimensional network; wherein, the unit latitude is set to 1°. Based on the altitude range corresponding to the first target area, the first target area is vertically divided into multiple independent geographical units, with each unit altitude as the basis; wherein, the unit altitude is set to 200m. Based on the coordinates corresponding to the geographic unit, the latitude range and altitude range corresponding to the geographic unit are written into the first target geographic element file to obtain the second target geographic element file; wherein, the first target geographic element file uses the same spatial coordinate system as the preprocessed sample satellite remote sensing image data; For any given geographic unit, the target crop sample area located within the latitude and altitude range corresponding to the geographic unit is determined based on the second target geographic feature file. The crop sample area is then constructed by combining the tiled multispectral image, latitude data, altitude data, slope data, data sampling time, and corresponding crop label data. This yields the crop sample data corresponding to the geographic unit. The crop label data consists of real crop labels input through human-computer interaction. Traverse all geographic units to obtain crop sample data corresponding to each geographic unit.

5. The cross-regional crop remote sensing image processing method integrating geographic information according to claim 4, characterized in that, After obtaining multiple geographic units and the crop sample data corresponding to each geographic unit, the method further includes: performing a time consistency check on the crop sample data, identifying the crop sample data that failed the check, and marking the crop sample data that failed the check.

6. The cross-regional crop remote sensing image processing method integrating geographic information according to claim 4, characterized in that, Based on the target geographic unit and its corresponding target geographic prototype library, a crop classification prediction model is trained, including: Based on the target crop sample data in the target geographic prototype library corresponding to the target geographic unit, extract the latitude data, altitude data, slope data, and the latitude vector, altitude vector, slope vector, and time vector corresponding to the data sampling time from the target crop sample data. The multispectral images of the target crop sample data are stitched together with latitude vector, altitude vector, slope vector and time vector to obtain sample fusion feature data; Using the sample fusion feature data as the actual input and the crop label data in the target crop sample data as the expected output, a crop classification prediction model is trained.

7. The cross-regional crop remote sensing image processing method integrating geographic information according to claim 6, characterized in that, Real-time geographic information data and real-time satellite remote sensing image data corresponding to the second target area are collected. A crop classification prediction model is used to identify the real-time geographic information data and real-time satellite remote sensing image data to obtain crop classification prediction results, including: Collect real-time geographic information data and real-time satellite remote sensing image data corresponding to the second target area; Based on the real-time geographic information data and real-time satellite remote sensing image data, determine the real-time slice multispectral image, real-time latitude data, real-time altitude data, real-time slope data, and real-time data sampling time corresponding to the second target area; Based on the real-time sliced ​​multispectral image, real-time latitude data, real-time altitude data, real-time slope data, and real-time data sampling time, real-time fused feature data is obtained; The crop classification prediction model is used to identify the real-time fused feature data to obtain crop classification prediction results.

8. A cross-regional crop remote sensing image processing apparatus that integrates geographic information, the apparatus being used to execute the cross-regional crop remote sensing image processing method that integrates geographic information as described in any one of claims 1 to 7, characterized in that... include: The sample data preprocessing module is used to acquire sample geographic information data and sample satellite remote sensing image data corresponding to the first target area, and to preprocess the sample geographic information data and sample satellite remote sensing image data to obtain preprocessed data; wherein, the first target area represents the region where data needs to be collected. The prototype library construction module is used to divide geographical units and bind data according to the preprocessed data to obtain multiple geographical units and crop sample data corresponding to each geographical unit, and to build a geographical prototype library corresponding to the geographical unit based on the crop sample data corresponding to the geographical unit. The prediction instruction response module is used to respond to a real-time crop prediction instruction input by the user through human-computer interaction, and to determine the target geographic unit corresponding to the second target area and its corresponding target geographic prototype library; wherein, the second target area represents the area determined in the first target area according to the real-time crop prediction instruction; The crop classification prediction module is used to train a crop classification prediction model based on the target geographic unit and its corresponding target geographic prototype library, and to collect real-time geographic information data and real-time satellite remote sensing image data corresponding to the second target area. The crop classification prediction model is then used to identify the real-time geographic information data and real-time satellite remote sensing image data to obtain crop classification prediction results.

9. An electronic device, characterized in that, It includes a memory and a processor, the memory being used to store a set of computer instructions; when the processor executes the set of computer instructions, it performs the operational steps of the method according to any one of claims 1 to 7.

Images