Farmland parcel identification method, device and equipment based on multi-source data

Abstract

The invention provides a farmland plot identification method, device and equipment based on multi-source data, and the method comprises the steps: obtaining first remote sensing data and second remote sensing data of a research region; wherein the resolution of the second remote sensing data is higher than that of the first remote sensing data; based on the first remote sensing data and a first target model, dividing the research area into at least one type of first agricultural plots; the at least one type of target area represents land parcel distribution of different land utilization types of the research area; based on the at least one type of target area, the second remote sensing data and a second target model, generating a target type block corresponding to the research area; the target category blocks represent land parcel distribution of the target land utilization type in the research area. By adopting the method, the precision of agricultural plot identification is improved.

Description

Technical Field

[0001] This application relates to, but is not limited to, the field of image recognition technology, and particularly relates to a method, device, and equipment for identifying farmland plots based on multi-source data. Background Art

[0002] Currently, in actual agricultural information service applications, the mainstream methods for obtaining agricultural plot information include on-site measurement surveys and visual interpretation by professionals based on high-resolution images. This process consumes a lot of manpower and material resources, has a high cost, and is difficult to update. With the significant progress of Convolutional Neural Networks (CNN) in the fields of image target detection and scene classification, more and more researchers have begun to apply CNN to the semantic segmentation task of remote sensing images. Since the input image needs to be adjusted to a fixed size when processed by CNN, and the contextual information between pixels in the image cannot be effectively utilized, the accuracy of the identified farmland plots is reduced. Summary of the Invention

[0003] In view of this, at least one method, device, and equipment for identifying farmland plots based on multi-source data are provided in the embodiments of this application.

[0004] The technical solution of the embodiments of this application is implemented as follows:

[0005] In a first aspect, an embodiment of this application provides a method for identifying farmland plots based on multi-source data, including: obtaining first remote sensing data and second remote sensing data of a research area; where the resolution of the second remote sensing data is higher than that of the first remote sensing data; dividing the research area into at least one type of target area based on the first remote sensing data and a first target model; at least one type of target area represents the plot distribution of different land use types in the research area; generating a target category block corresponding to the research area based on at least one type of target area, the second remote sensing data, and a second target model; the target category block represents the plot distribution of the target land use type in the research area.

[0006] In a second aspect, an embodiment of this application provides a device for identifying farmland plots based on multi-source data, including: an acquisition module, configured to obtain first remote sensing data and second remote sensing data of a research area; where the resolution of the second remote sensing data is higher than that of the first remote sensing data; a division module, configured to divide the research area into at least one type of target area based on the first remote sensing data and a first target model; at least one type of target area represents the plot distribution of different land use types in the research area; a generation module, configured to generate a target category block corresponding to the research area based on at least one type of target area, the second remote sensing data, and a second target model; the target category block represents the plot distribution of the target land use type in the research area.

[0007] In a third aspect, an embodiment of the present application provides an electronic device, including a memory and a processor. The memory stores a computer program that can run on the processor, and when the processor executes the program, some or all of the steps in the above method are implemented.

[0008] A method, apparatus, and device for identifying farmland plots based on multi-source data provided by an embodiment of the present application divide a research area into at least one type of land use plot through remote sensing data with a first resolution, so as to reduce the problem of missed identification caused by incomplete remote sensing data. Through the classification and identification of the research area, the workload of subsequent extraction of target type plots can be reduced. On the basis of dividing the research area into at least one type of land use plot, the remote sensing data with a second resolution of the research area is combined to identify the target categories in the research area block by block, improving the accuracy of identifying the plots in the research area.

[0009] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and do not limit the technical solution of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The accompanying drawings herein are incorporated into the specification and constitute a part of this specification. These drawings illustrate embodiments consistent with the present application and, together with the specification, are used to explain the technical solution of the present application.

[0011] Figure 1 It is a schematic flow chart of the implementation of a method for identifying farmland plots based on multi-source data provided by an embodiment of the present application;

[0012] Figure 2 It is a schematic flow chart of the implementation of a method for identifying farmland plots based on multi-source data provided by an embodiment of the present application;

[0013] Figure 3 It is a schematic flow chart of the implementation of a method for identifying farmland plots based on multi-source data provided by an embodiment of the present application;

[0014] Figure 4 It is a schematic flow chart of the implementation of a method for identifying farmland plots based on multi-source data provided by an embodiment of the present application;

[0015] Figure 5 It is a schematic flow chart of the implementation of a method for identifying farmland plots based on multi-source data provided by an embodiment of the present application;

[0016] Figure 6 It is a schematic flow chart of the implementation of a method for identifying farmland plots based on multi-source data provided by an embodiment of the present application;

[0017] Figure 7Schematic diagram of the implementation process of a farmland plot recognition method based on multi-source data provided by an embodiment of the present application;

[0018] Figure 8 Schematic diagram of the implementation process of a farmland plot recognition method based on multi-source data provided by an embodiment of the present application;

[0019] Figure 9 Schematic diagram of the implementation process of a crop classification method provided by an embodiment of the present application;

[0020] Figure 10 Schematic diagram of the implementation process of a farmland plot recognition method based on multi-source data provided by an embodiment of the present application;

[0021] Figure 11 Schematic diagram of the cultivated land type provided by an embodiment of the present application;

[0022] Figure 12 Schematic diagram of the composition structure of a farmland plot recognition device based on multi-source data provided by an embodiment of the present application;

[0023] Figure 13 Schematic diagram of the hardware entity of an electronic device provided by an embodiment of the present application. Detailed implementation manners

[0024] In order to make the objectives, technical solutions, and advantages of the present application clearer, the technical solutions of the present application will be further elaborated in detail below with reference to the accompanying drawings and embodiments. The described embodiments should not be construed as limitations on the present application. All other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present application.

[0025] In the following descriptions, reference is made to "some embodiments", which describe subsets of all possible embodiments. However, it can be understood that "some embodiments" can be the same subsets or different subsets of all possible embodiments, and can be combined with each other without conflict. The terms "first / second / third" involved are only used to distinguish similar objects and do not represent a specific order for the objects. It can be understood that "first / second / third" can be interchanged with a specific order or sequence when allowed, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0026] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which this application belongs. The terms used herein are only for the purpose of describing this application and are not intended to limit this application.

[0027] At present, in the actual application of agricultural information services, the mainstream methods for obtaining agricultural plot information include on-site measurement surveys and visual interpretation by professionals based on high-resolution images. This process consumes a lot of manpower and material resources, with high costs and difficulties in updating. With the significant progress of Convolutional Neural Networks (CNNs) in fields such as image object detection and scene classification, more and more researchers have started to apply CNNs to the semantic segmentation task of remote sensing images. Since the input image needs to be adjusted to a fixed size when processed by CNNs, and the contextual information between pixels in the image cannot be effectively utilized, this has led to a certain degree of decline in classification accuracy.

[0028] The embodiments of the present application provide a method for identifying farmland plots based on multi-source data, which can be executed by a processor of an electronic device. Among them, the electronic device can refer to devices with data processing capabilities such as servers, laptop computers, tablet computers, desktop computers, smart TVs, set-top boxes, mobile devices (such as mobile phones, portable video players, personal digital assistants, dedicated messaging devices, portable game devices), etc.

[0029] Figure 1 FIG. is a schematic flowchart of the implementation of a method for identifying farmland plots based on multi-source data provided by the embodiments of the present application, which can be executed by a processor of an electronic device. As Figure 1 shown, the method includes the following steps S101 to S103, which will be described in combination with Figure 1 the steps presented.

[0030] Step S101: Obtain the first remote sensing data and the second remote sensing data of the research area.

[0031] Among them, the resolution of the second remote sensing data is higher than that of the first remote sensing data.

[0032] In some embodiments, the first remote sensing data and the second remote sensing data of the research area can be obtained from public data platforms (such as commercial remote sensing satellite data suppliers, government agency data platforms, resource satellite application centers, etc.), and can also be obtained through satellites (such as Sentinel series satellites). Among them, the first remote sensing data can include optical satellite remote sensing data and SAR satellite remote sensing data.

[0033] In some embodiments, the resolution of the first remote sensing data can be medium-resolution remote sensing data such as 10 meters, 15 meters, etc.; the resolution of the second remote sensing data can be high-resolution remote sensing data such as 0.5 meters, 0.8 meters, etc.

[0034] In some embodiments, after obtaining the first remote sensing data and the second remote sensing data of the study area, it is also necessary to preprocess the first remote sensing data and the second remote sensing data, including: for the first remote sensing data, geometric correction, radiometric correction, cloud detection and removal, position matching between multi-temporal images, and terrain flattening and denoising processing of SAR data are required to obtain the preprocessed first remote sensing data; for the second remote sensing data, after cloud removal processing, it is recommended to screen the second remote sensing data through preliminary preprocessing experiments and the like to obtain the second remote sensing data that is more effective for land use classification.

[0035] Step S102: Based on the first remote sensing data and the first target model, divide the study area into at least one type of target area.

[0036] Wherein, the at least one type of target area represents the plot distribution of different land use types in the study area.

[0037] In some embodiments, the first target model is a trained random forest model, wherein the trained random forest model is trained based on the publicly available European Space Agency (ESA) series dataset and the China Land Cover Dataset (CLCD); the training process of the trained random forest model includes: obtaining remote sensing data of multiple study areas from the publicly available European Space Agency dataset and the China Land Cover Dataset as the training set; inputting the remote sensing data of multiple study areas into the random forest model to be trained to obtain at least one predicted block of multiple study areas, and based on the loss value between the at least one predicted block and the true block of the study area, adjusting the model parameters of the random forest model to be trained until the convergence function is satisfied, and then outputting the trained random forest model.

[0038] In some embodiments, the at least one type of target area may represent land use types such as cultivated land, forest and grassland, built-up area, water body area, and others in the study area.

[0039] In some embodiments, input the first remote sensing data of the study area into the first target model to obtain at least one type of target area of the study area; for example, input the medium-resolution remote sensing data of the study area into the trained random forest model to divide the study area into land use types such as cultivated land, forest and grassland, built-up area, water body area, and others.

[0040] Step S103: Based on the at least one type of target area, the second remote sensing data, and the second target model, generate the target category blocks corresponding to the study area.

[0041] Among them, the target category blocks represent the plot distribution of the target land use type in the research area.

[0042] In some embodiments, the second target model is a semantic segmentation model, such as a Fully-Convolutional Network (FCN), a U-Net, a DeepLab, a Mask Region Convolutional Neural Network (Mask R-CNN), etc.

[0043] In some embodiments, the target category blocks may be cultivated land plots, forest and grass plots, built-up area plots, etc. in the research area.

[0044] In some embodiments, at least one type of target area and the second remote sensing data of the research area are input into the second target model to obtain the target category blocks in the research area; for example, the research area divided into cultivated land, forest and grassland, built-up area, water area, others and the high-resolution remote sensing data are input into the trained fully convolutional network to obtain the cultivated land type plots in the research area.

[0045] In the embodiments of the present application, the research area is divided into at least one type of land use plot by medium-resolution remote sensing data to reduce the problem of missed recognition caused by incomplete remote sensing data. Through the classification and recognition of the research area, the workload of subsequent extraction of target type plots can be reduced. On the basis of dividing the research area into at least one type of land use plot, the high-resolution remote sensing data of the research area is combined to identify the target category blocks in the research area, so as to improve the accuracy of identifying the plots in the research area.

[0046] Figure 2 Schematic diagram of the implementation process of a method for identifying farmland plots based on multi-source data provided by the embodiments of the present application. This method can be executed by the processor of an electronic device. Based on Figure 1 , Figure 1 The step S103 in can be updated to step S201 to step S202, and will be described in combination with Figure 2 the steps shown.

[0047] Step S201: Generate at least one type of target plot corresponding to the research area based on the at least one type of target area, the second remote sensing data, and the second target model.

[0048] In some embodiments, the at least one type of target plot may be regular cultivated land plots, terraced cultivated land plots, sloping cultivated land plots, forest-interspersed cultivated land plots, etc. in the cultivated land plots; it may also be urban built-up plots, rural built-up plots, factory built-up plots, etc. in the built-up area.

[0049] In some embodiments, the second remote sensing data of the at least one type of target area and the research area is input into a second target model to obtain at least one type of target plot in the research area; for example, the research area divided into cultivated land, forest and grassland, built-up area, water area, others, and high-resolution remote sensing data are input into a trained U-shaped network to generate regular cultivated land plots, terraced cultivated land plots, sloping cultivated land plots, forest-interspersed cultivated land plots, etc. in the cultivated land plots of the area.

[0050] Step S202: Generate a target category block corresponding to the research area based on the at least one type of target plot.

[0051] In some embodiments, the at least one type of target plot is spliced to obtain a target type block corresponding to the research area; it can be understood that the area ranges of each type of target plot in the at least one type of target plot are spliced in the area range of the research area to obtain the distribution of the target category block in the research area. For example, the area ranges of regular cultivated land plots, terraced cultivated land plots, sloping cultivated land plots, and forest-interspersed cultivated land plots in the cultivated land plots are spliced in the area range of the research area to obtain the distribution of different types of cultivated land plots in the research area.

[0052] In the embodiments of the present application, by dividing the research area into at least one type of target area and high-resolution remote sensing data of the research area, at least one type of target plot of the land use type to be studied is obtained; the at least one target plot is fused, spliced, etc. to obtain the block of the target category to be studied in the research area.

[0053] Figure 3 It is a schematic flowchart of the implementation process of a method for identifying farmland plots based on multi-source data provided by the embodiments of the present application, and this method can be executed by the processor of an electronic device. Based on Figure 2 , the at least one type of target plot includes regular cultivated land plots; the second target model includes a first semantic segmentation model. Figure 2 Step S201 in Figure 3 can be updated to steps S301 to S305, and will be described in combination with the steps shown in

[0054] Step S301: Based on the second remote sensing data, obtain a first target object for cultivated land area segmentation in the first area.

[0055] Among them, the first area includes the area in the research area with a slope less than the first preset slope.

[0056] In some embodiments, it is first necessary to obtain the elevation data of the research area, including the slope of each position in the research area.

[0057] In some embodiments, based on the slope of each location in the study area, an area with a slope less than a first preset slope is taken as a first area; wherein the first area may be a flat area in the study area, and the first preset slope may be 12°; exemplarily, an area in the study area with a slope less than 12° is taken as a flat area.

[0058] In some embodiments, the first target object for segmenting the cultivated land area refers to an object for segmenting the cultivated land parcels in the first area, such as roads and water systems interspersed in the cultivated land parcels.

[0059] For example, in high-resolution remote sensing data, areas such as roads and water systems in flat areas are obtained.

[0060] In some embodiments, the at least one type of target area may also include built-up areas and water bodies.

[0061] In some embodiments, after acquiring the first area, it is also necessary to eliminate the areas corresponding to the built-up areas and water bodies in the first area; it can be understood that in the first remote sensing data, the layer in the study area where the layer of the first area is located is determined, and the layer of the built-up area and water body in the study area is determined, the layer of the first area and the layer of the built-up area and water body in the study area are superimposed to determine the layer of the built-up area and water body in the first area, and the layer of the built-up area and water body in the first area is eliminated, and based on the layer of the first area after eliminating the layer of the built-up area and water body, the first area of ​​the area where the built-up area and water body are eliminated is determined.

[0062] Step S302: segment the first region based on the first target object to obtain at least one first sub-region.

[0063] In some embodiments, it can be understood that the first area is divided into a plurality of first sub-areas by interspersing the areas of roads, water systems, etc. in the cultivated land plots of the first area. Exemplarily, a layer of roads, water systems, etc. is obtained in the high-resolution remote sensing data, and the layer of roads, water systems, etc. in the high-resolution remote sensing data is superimposed on the layer of the flat dam area of ​​the medium-resolution remote sensing data, and the layer of roads, water systems, etc. interspersed on the layer of the first area is divided into at least one first sub-layer, wherein each first sub-layer corresponds to a first sub-area in the first area.

[0064] In some embodiments, after obtaining at least one first sub-region, edge detection can be performed on each sub-region based on an edge detection model to obtain an edge strength map of each first sub-region, wherein the edge detection model can be a Dense Extreme Inception Network for Edge Detection (DexiNed).

[0065] Step S303: Based on the second remote sensing data, obtain at least one first suspected regular cultivated land plot in the at least one first sub-region through a vectorization method.

[0066] In some embodiments, in the second remote sensing data, vectorize the grids where the at least one first sub-region is located, and screen the vectorized grids where the at least one first sub-region is located based on the grid bands that can represent regular cultivated land, so as to obtain the first suspected regular cultivated land plots in the first region.

[0067] Step S304: Based on the at least one first sub-region, the first semantic segmentation model, and the second remote sensing data, obtain at least one second suspected regular cultivated land plot in the at least one first sub-region.

[0068] In some embodiments, the first semantic segmentation model may be a U-Net++ model.

[0069] In some embodiments, input the at least one first sub-region and the second remote sensing data of the study area into the first semantic model to obtain a segmentation result indicating whether each first sub-region is a regular cultivated land plot, and obtain the second suspected regular cultivated land blocks based on the segmentation result.

[0070] Exemplarily, input the at least one first sub-region and the high-resolution remote sensing data into the trained U-Net++ model to output a segmentation result indicating whether each first sub-region is a regular cultivated land, and generate a second suspected regular cultivated land plot based on the segmentation result of each first sub-region.

[0071] Step S305: Overlay the first suspected regular cultivated land plots and the second suspected regular cultivated land plots of the at least one first sub-region to generate the regular cultivated land plots.

[0072] In some embodiments, take the intersection of the first suspected regular cultivated land plots and the second suspected regular cultivated land of the at least one first sub-region to obtain the regular cultivated land blocks in the study area.

[0073] Exemplarily, in the medium-resolution remote sensing data, overlay the layers of the first suspected regular cultivated land plots and the second suspected regular cultivated land plots, use the repeated layers as the layers of the regular cultivated land plots, and determine the regular cultivated land plots in the study area based on the layers of the regular cultivated land plots.

[0074] In an embodiment of the present application, based on the distribution pattern of regular farmland, regular farmland plots are extracted in the flat area of ​​the study area, reducing the subsequent extraction work; after the flat area is divided into blocks based on the areas corresponding to the water system and the roads, the first suspected regular farmland plot in the study area is obtained based on the high-resolution remote sensing data, and the second suspected regular farmland plot in the study area is obtained based on the high-resolution remote sensing data. The intersection of the first suspected regular farmland plot obtained by the high-resolution remote sensing data and the second suspected regular farmland plot obtained by the high-resolution remote sensing data is taken to obtain the accuracy of the extracted regular farmland plots.

[0075] Figure 4 This is a schematic diagram of an implementation flow of a method for identifying farmland plots based on multi-source data provided in an embodiment of the present application. The method can be executed by a processor of an electronic device. Figure 2 , the at least one type of target plot includes forest-cultivated land plots; the at least one type of target area includes forest-grass area and cultivated land area; the second target model includes a second semantic segmentation model; Figure 2 Step S201 in can be updated to step S401 to step S404, combining Figure 4 The steps shown are explained.

[0076] Step S401: Acquire a second target object in the target forest and grassland area for segmenting the cultivated land area.

[0077] Among them, the target forest and grassland area represents the area in the forest and grassland area where the forest and grassland area reaches a preset threshold.

[0078] In some embodiments, obtaining the target forest and grassland area includes: dividing the study area into at least one target forest and grassland area, dividing the forest and grassland area into 100-meter grids, and determining all grids with a forest and grassland area greater than 80 meters as the target forest and grassland area, which can be determined based on the following formula (1).

[0079]

[0080] Among them, R is the area ratio of forest to grassland in the grid, Num 林草地 is the number of pixels in the grid that are forest or grassland.

[0081] In some embodiments, the second target object for segmenting the cultivated land area refers to the object for segmenting the cultivated land parcels in the target forest and grassland parcels, such as roads and water systems interspersed in the cultivated land parcels.

[0082] Step S402: segment the target forest and grassland block based on the second target object to obtain at least one second sub-region.

[0083] In some embodiments, the first region is divided into a plurality of second sub-regions by areas such as roads and water systems interspersed in the cultivated land plots of the target forest and grassland. Exemplarily, layers of roads, water systems, etc. are obtained from high-resolution remote sensing data, and the layers of roads, water systems, etc. in the high-resolution remote sensing data are overlaid on the layer of the target forest and grassland plots in the medium-resolution remote sensing data. The layer of the target forest and grassland plots is divided into at least one second sub-layer by the layers of roads, water systems, etc. interspersed on the layer of the target forest and grassland plots, where each second sub-layer corresponds to a second sub-region in the target forest and grassland plots.

[0084] Step S403: Generate suspected forested cultivated land plots in the at least one second sub-region based on the at least one second sub-region, the second remote sensing data, and the second semantic segmentation model.

[0085] In some embodiments, the second semantic segmentation model may be a Deeplab v3+ model.

[0086] In some embodiments, the at least one second sub-region and the second remote sensing data of the study region are input into the second semantic segmentation model to obtain a segmentation result indicating whether each second sub-region is a forested cultivated land plot. Based on the segmentation result for each second sub-region, suspected forested cultivated land plots in the at least one second sub-region are obtained.

[0087] Exemplarily, the at least one second sub-region and the high-resolution remote sensing data are input into the trained Deeplabv3+ model to generate a segmentation result indicating whether each second sub-region is a forested cultivated land plot. Based on the segmentation result for each second sub-region, suspected forested cultivated land plots in the at least one second sub-region are obtained.

[0088] Step S404: Overlay the suspected forested cultivated land plots and the cultivated land plots to generate the forested cultivated land plots.

[0089] In some embodiments, the suspected forested cultivated land plots in the at least one second sub-region and the cultivated land plots in at least one target region are overlaid to obtain regular cultivated land blocks in the study region.

[0090] Exemplarily, in the medium-resolution remote sensing data, the layer of the suspected forested cultivated land plots and the layer of the cultivated land area in at least one target region are overlaid, and the overlapping layers are used as the layer of the regular cultivated land plots. Based on the position of the layer of the regular cultivated land area in the medium-resolution remote sensing data of the study region, the regular cultivated land plots in the study region are determined.

[0091] In the embodiments of the present application, based on the distribution law of forested cultivated land, in the forest and grass areas of the research area, forest and grass areas with a forest and grass ratio greater than 80% are obtained as target forest and grass areas, reducing the workload of extracting forested cultivated land while improving the accuracy of subsequent extraction; based on areas such as roads and water systems in the target forest and grass areas, the forest and grass areas are divided into multiple second sub-areas to identify the high-resolution remote sensing data of each second sub-area to obtain whether each second sub-area is a suspected forested cultivated land, and then combined with the cultivated land areas of the research area to obtain forested cultivated land plots with high accuracy, thereby improving the accuracy of obtaining forested cultivated land.

[0092] Figure 5 It is a schematic flowchart of the implementation of a method for identifying farmland plots based on multi-source data provided by the embodiments of the present application, and this method can be executed by the processor of an electronic device. Please refer to Figure 2 , the at least one type of target plot includes terraced cultivated land plots; the second target model includes a third semantic segmentation model; Figure 2 The step S201 in Figure 5 will be described with reference to the steps shown.

[0093] Step S501: Screen the second area in the research area to obtain a first screening area.

[0094] Among them, the second area includes the area in the research area with a slope greater than a first preset slope; the first screening area includes the area in the second area with a slope greater than the first preset slope and less than or equal to a second preset slope.

[0095] In some embodiments, it is first necessary to obtain the elevation data of the research area, including the slope of each position in the research area.

[0096] In some embodiments, based on the slope of each position in the research area, the area with a slope greater than the first preset slope is used as the second area; among them, the second area can be the mountainous area in the research area, and the first preset slope can be 12°; for example, the area in the research area with a slope greater than 12° is used as the mountainous area.

[0097] In some embodiments, based on the slope of each position in the second area, the area with a slope greater than the first preset slope and less than the second preset slope is used as the first screening area; among them, the second preset slope can be 25°; for example, the area in the second area with a slope greater than 12° and less than 25° is used as the first screening area.

[0098] Step S502: Obtain the third target object for dividing the cultivated land area in the first screening area.

[0099] In some embodiments, the third target object for the cultivated area segmentation refers to the object that segments the cultivated land plots in the second area, such as roads, water systems, etc. interspersed in the cultivated land plots, as well as ridge lines, valley lines, etc. in the second area obtained based on elevation data.

[0100] Step S503: Segment the first screening area based on the third target object to obtain at least one third sub-area.

[0101] In some embodiments, it can be understood that the first screening area is divided into multiple third sub-areas through the areas where roads, water systems, ridge lines, valley lines, etc. interspersed in the first screening area. Exemplarily, layers of roads, water systems, ridge lines, valley lines, etc. are obtained from high-resolution remote sensing data, and the layers of roads, water systems, ridge lines, valley lines, etc. in the high-resolution remote sensing data are overlaid on the layer of the first screening area in the medium-resolution remote sensing data. The layer of the first screening area is divided into at least one third sub-layer through the layers of roads, water systems, ridge lines, valley lines, etc. interspersed on the layer of the first screening area, where each third sub-layer corresponds to a third sub-area in the first screening area.

[0102] Step S504: Based on the second remote sensing data, obtain the first suspected terraced cultivated land plots in the at least one third sub-area through a vectorization method.

[0103] In some embodiments, first, the fine boundaries need to be extracted based on an edge detection model to obtain the edge intensity map of each third sub-area. Among them, the edge detection model can be the Rich Feature Hierarchy for Edge Detection (RCF).

[0104] In some embodiments, in the second remote sensing data, the grids where at least one third sub-area is located are vectorized to obtain the vectorized grids of at least one third sub-area; the vectorized grid range where at least one third sub-area is located is screened based on the grid bands that can represent terraced cultivated land to obtain the grid range of the first suspected regular cultivated land plots in the first screening area, and the first suspected regular cultivated land plots in the first screening area are determined based on the grid range of the first suspected regular cultivated land plots.

[0105] Step S505: Based on the at least one third sub-area, the second remote sensing data, and the third semantic segmentation model, obtain the second suspected terraced cultivated land plots in the at least one third sub-area.

[0106] In some embodiments, the third semantic segmentation model can be the U-net++ model.

[0107] In some embodiments, at least one third sub-region and second remote sensing data of the research region are input into a third semantic segmentation model to obtain a segmentation result indicating whether each third sub-region is a terraced cultivated land plot. Based on the segmentation result of each third sub-region, a second suspected terraced cultivated land plot is generated.

[0108] Exemplarily, at least one third sub-region and high-resolution remote sensing data of the research region are input into a trained U-Net++ model to obtain a segmentation result indicating whether each third sub-region is a terraced cultivated land plot. The third sub-regions characterized as terraced cultivated land plots are merged to obtain a second suspected terraced cultivated land plot.

[0109] Step S506: Superimpose the first suspected terraced cultivated land plot and the second suspected terraced cultivated land plot of the at least one third sub-region to generate the terraced cultivated land plot.

[0110] In some embodiments, the intersection of the first suspected terraced cultivated land plot and the second suspected terraced cultivated land of at least one third sub-region is taken to obtain regular cultivated land blocks in the research region.

[0111] In some embodiments, exemplarily, in the medium-resolution remote sensing data, the layer of the first suspected terraced cultivated land plot and the layer of the second suspected terraced cultivated land plot are superimposed, and the overlapping layers are used as the layer of the terraced cultivated land plot. Based on the position of the layer of the terraced cultivated land plot in the remote sensing data of the research region, the regular cultivated land plots in the research region are determined.

[0112] In the embodiments of the present application, based on the distribution law of terraced cultivated land, in the mountainous area of the research region, a first screening region with a slope greater than 12° and less than 25° is obtained to obtain terraced cultivated land blocks from the first screening region, reducing the workload of extracting terraced cultivated land in the research region. The first screening region is segmented into multiple third sub-regions based on regions such as water bodies, roads, valley lines, and ridge lines in the first screening region. In the high-resolution remote sensing data, a first suspected terraced cultivated land plot of multiple third sub-regions is obtained based on the vectorization method; a second suspected terraced cultivated land plot is obtained through the high-resolution remote sensing data of the third sub-region, and the terraced cultivated land plot in the research region is obtained based on the first suspected terraced cultivated land plot and the second suspected terraced cultivated land plot, combining high-resolution remote sensing data and medium-resolution remote sensing data, and improving the accuracy of identifying terraced cultivated land plots in the research region.

[0113] Figure 6 It is a schematic implementation flowchart of a method for identifying farmland plots based on multi-source data provided by the embodiments of the present application, and this method can be executed by a processor of an electronic device. Please refer to Figure 2 , the at least one type of target plot includes sloping cultivated land plots; the second target model includes a fourth semantic segmentation model;Figure 2 Step S201 therein can be updated to Step S601 to Step S604, which will be described in conjunction with the steps Figure 6 shown.

[0114] Step S601: Screen the second area in the research area to obtain a second screening area.

[0115] Wherein, the second area includes the area in the research area with a slope greater than a first preset slope.

[0116] Wherein, the second screening area includes the area in the second area with a slope greater than a second preset slope; the second preset slope is greater than the first preset slope.

[0117] In some embodiments, based on the slope of each position in the elevation data of the research area, the area with a slope greater than the first preset slope is used as the second area; wherein, the second area can be the mountainous area in the research area, and the first preset slope can be 12°; for example, the area in the research area with a slope greater than 12° is used as the mountainous area.

[0118] In some embodiments, based on the slope of each position in the second area, the area with a slope greater than the second preset slope is used as the second screening area; wherein, the second preset slope can be 25°; for example, the area in the second area with a slope greater than 25° is used as the second screening area.

[0119] Step S602: Obtain a fourth target object for dividing the cultivated land area in the second screening area.

[0120] In some embodiments, the fourth target object for dividing the cultivated land area refers to the object for dividing the cultivated land plots in the second area, such as roads, water systems, etc. interspersed in the cultivated land plots, and the ridge lines, valley lines, etc. in the second area obtained based on the elevation data.

[0121] Step S603: Divide the second screening area based on the fourth target object to obtain at least one fourth sub-area.

[0122] In some embodiments, the second screening area is divided into a plurality of fourth sub-areas by areas where roads, water systems, ridge lines, valley lines, etc. are interspersed in the cultivated land plots in the second screening area. Exemplarily, layers of roads, water systems, ridge lines, valley lines, etc. are obtained from high-resolution remote sensing data, and the layers of roads, water systems, ridge lines, valley lines, etc. in the high-resolution remote sensing data are overlaid on the layer of the second screening area in the medium-resolution remote sensing data. The layer of the second screening area is divided into at least one fourth sub-layer by the layers of roads, water systems, ridge lines, valley lines, etc. interspersed in the layer of the second screening area, where each fourth sub-layer corresponds to a fourth sub-area in the second screening area.

[0123] Step S604, generate the sloping cultivated land plots based on the at least one fourth sub-area, the second remote sensing data, and the fourth semantic segmentation model.

[0124] In some embodiments, the fourth semantic segmentation model may be a U-Net++ model.

[0125] In some embodiments, the at least one fourth sub-area and the second remote sensing data of the study area are input into the fourth semantic segmentation model to obtain a segmentation result indicating whether each fourth sub-area is a sloping cultivated land plot. Based on the segmentation result of each fourth sub-area, sloping cultivated land plots are generated.

[0126] Exemplarily, the at least one fourth sub-area and the high-resolution remote sensing data of the study area are input into the trained U-Net++ model to obtain a segmentation result indicating whether each fourth sub-area is a sloping cultivated land plot. The fourth sub-areas characterized as terraced cultivated land plots are merged to obtain the sloping cultivated land plots of the study area.

[0127] In the embodiments of the present application, based on the distribution law of sloping cultivated land, in the mountainous area of the study area, the second screening area with a slope greater than 25° is obtained to obtain the sloping cultivated land blocks from the second screening area, reducing the workload of extracting sloping cultivated land in the study area. The second screening area is segmented into a plurality of fourth sub-areas based on the areas of water bodies, roads, valley lines, ridge lines, etc. in the second screening area, and the sloping cultivated land plots are obtained through the high-resolution remote sensing data of the fourth sub-areas, improving the accuracy of identifying sloping cultivated land plots in the study area.

[0128] Figure 7 It is a schematic flowchart of the implementation of a method for identifying farmland plots based on multi-source data provided by the embodiments of the present application, and this method can be executed by the processor of an electronic device. Based on Figure 1 , Figure 1 Step S102 in can be updated to steps S701 to S702, and will be described in combination with the steps shown in Figure 7 shown.

[0129] Step S701: Generate target feature data of the research area based on the first remote sensing data.

[0130] In some embodiments, the first remote sensing data may include optical data, the target feature data may include index features of the research area, the spectral features are determined based on the optical data, and the index features are determined based on the spectral features.

[0131] In some embodiments, the spectral features include the red band (Red), green band (Green), blue band (Blue), near-infrared band (Nir), short-wave infrared 1 band (SWIR1), and short-wave infrared 2 (SWIR2) band, and the index features include the normalized difference vegetation index (NVDI), normalized difference senescent vegetation index (NDSVI), land surface water index (LSWI), modified normalized difference water index (MNDWI), normalized difference yellow index (NDYI), and normalized difference soil index (NDSI). The index features are calculated based on the spectral features. Please refer to Table 1 below.

[0132] Table 1

[0133]

[0134] In some embodiments, the first remote sensing data further includes SAR data, the target feature data may include SAR features, and the SAR features are determined based on the SAR data.

[0135] In some embodiments, the SAR features include that the radar emits vertically polarized waves and receives horizontally polarized waves (VerticalHorizontal, VH), and the radar emits vertically polarized waves and simultaneously receives vertically polarized waves (Vertical Vertical, VV), and the cross-polarization ratio (Cross-Polarization Ratio, CR). Among them, the VV polarization and VH polarization can be directly obtained based on the SAR data, and the cross-polarization ratio can be obtained through the following formula (2).

[0136]

[0137] Wherein, and Table shows the backscattering coefficients of VH and VV in the logarithmic domain.

[0138] Step S702: Generate at least one target area of the research area based on the target feature data and the first target model.

[0139] In some embodiments, the target features are input into the trained first target model to obtain at least one target area of the research area.

[0140] Exemplarily, the index features and SAR features of the research area are input into the trained random forest model to obtain at least one target area representing the land use type of the research area.

[0141] In the embodiments of the present application, target feature data of the research area is generated from the medium-resolution remote sensing data of the research area. Based on the target feature data, the medium-resolution remote sensing data, and the random forest model, the research area is divided into at least one target area based on the land use type, so as to reduce the problem of missing cultivated land plots in subsequent extraction, thereby improving the accuracy of plot recognition.

[0142] Figure 8 FIG. is a schematic implementation flowchart of a farmland plot recognition method based on multi-source data provided by the embodiments of the present application. This method can be executed by a processor of an electronic device. Based on Figure 7 , the first remote sensing data includes monthly optical data and monthly synthetic aperture radar (SAR) data; the target feature data includes target index features and target SAR features; Figure 7 The step S701 in Figure 8 can be updated to step S801 to step S802, and will be described in conjunction with

[0143] Step S801: Generate monthly index features based on the monthly optical data, and generate monthly SAR features based on the monthly SAR data.

[0144] In some embodiments, based on the method in the above step S701, index features and SAR features of one shooting cycle can be generated based on the first remote sensing data obtained in one satellite shooting cycle. Based on the indication features and SAR features obtained in multiple shooting cycles in a month, monthly index features and monthly SAR features can be obtained.

[0145] Step S802: Synthesize the monthly index features to obtain target index features, and synthesize the monthly SAR features to obtain target SAR features.

[0146] In some embodiments, the median synthesis method is used to synthesize the monthly index features to obtain target index features, and the median synthesis method is used to synthesize the monthly SAR features to obtain target SAR features.

[0147] In the embodiments of the present application, by generating monthly index features and monthly SAR features of the research area based on monthly medium-resolution remote sensing data of the research area, median synthesis is performed based on the monthly index features and monthly SAR features to obtain target feature data. The use of monthly medium-resolution remote sensing data enables the generated feature data to reflect the long-term land use trends of the long-term research area, improving the accuracy of the generated target feature data.

[0148] The following describes an exemplary application in an actual scenario of a farmland plot identification method provided by the embodiments of the present application based on multi-source data.

[0149] Cultivated land is the basic resource for human survival and development and the cornerstone of food security. With the improvement of social living standards and the development of agricultural modernization, the demands and structures of agricultural development are constantly changing, and cultivated land is facing double tests of quality and quantity. Rapidly and accurately extracting cultivated land area and distribution information is of great significance for agricultural production planning, planting structure adjustment, pest detection, important agricultural product production guarantee planning, etc., and is of great significance for ensuring food security and the development of agricultural modernization.

[0150] In the related art, the method of obtaining cultivated land distribution and area information through on-site surveys is time-consuming and laborious, consumes too much manpower and material resources, and cannot be updated in a timely manner. With the development of remote sensing observation technology, remote sensing image interpretation has become an important means of understanding cultivated land distribution. The extraction of cultivated land using medium- and low-resolution remote sensing images mainly adopts the method of land cover classification, which is used to quickly understand the macroscopic situation of cultivated land distribution and area on a large scale. Limited by conditions such as image resolution, it cannot meet the accuracy requirements of information products for actual agricultural production. With the launch and application of high-resolution remote sensing satellites, sub-meter high-resolution images have become the main data source for cultivated land extraction, which can well meet the refined requirements of plot extraction.

[0151] At present, in the actual application of agricultural information services, the mainstream method for obtaining plot information relies on professional personnel to visually interpret high-resolution images. This process consumes a lot of manpower and material resources, with high costs and difficulties in updating. With the significant progress of Convolutional Neural Networks (CNN) in fields such as image object detection and scene classification, more and more researchers have begun to apply CNN to the semantic segmentation task of remote sensing images. Deep learning technology has gradually been widely applied in the field of land cover classification of remote sensing images, such as the extraction of land object targets like buildings, roads, and water bodies. Since the input image needs to be adjusted to a fixed size when processed by CNN, and the contextual information between pixels in the image cannot be effectively utilized, this has led to a certain degree of decline in classification accuracy. To solve the above problems, some researchers have proposed a Fully Convolutional Networks (FCN) that replaces the fully connected layer with a convolutional layer, resulting in improved segmentation accuracy.

[0152] With the development of image edge detection technology, more and more models such as Holistically-nested Edge Detection (HED) and Richer Convolutional Features (RCF) have been proposed and achieved application effects similar to those of humans in many fields. In the field of cultivated land extraction using remote sensing images, relevant technologies have been migrated to the cultivated land extraction task through the DeepLabv3+ model, achieving the extraction of cultivated land from WorldView images with a resolution of 1m. The research results have proven that deep learning semantic segmentation technology can obtain higher-precision plot extraction results compared to traditional classification methods. Relevant technologies have also proposed building an edge detection model for remote sensing images (Full Dilated-RCF, FD-RCF) to extract the edges of cultivated land plots. Relevant technologies have also proposed cascading semantic segmentation models and edge detection models, and constructing an extraction model through boundary enhancement and focused training to extract cultivated land plots.

[0153] In the related art, a pixel-level crop classification method was proposed. According to the visual morphological characteristics presented by cultivated land in high-resolution remote sensing images, appropriate convolutional neural network models were respectively selected and improved to hierarchically extract various types of cultivated land, which not only avoided logical classification errors but also improved the classification efficiency. Finally, the extraction results of multiple cultivated land subtypes were fused and post-processed to obtain the extraction result map of the complete cultivated land morphology information. On the basis of the extraction of cultivated land morphology information, multi-temporal SAR data was used to construct plot-level time series features under boundary constraints. Then, by utilizing the classification ability of the recurrent neural network for time series data and combining field survey sampling data, the classification abilities of different SAR time series features and their combinations were analyzed and evaluated. Finally, the cultivation and utilization types of each plot were discriminated, so as to eliminate non-cultivated land and further optimize the extraction results. Please refer to Figure 9 , Figure 9 is a schematic implementation flow diagram of a crop classification method provided by an embodiment of the present application, including steps S901 to S906, which will be described in conjunction with Figure 9 the steps shown.

[0154] Step 901: Obtain high-spatial-resolution remote sensing images of the research area.

[0155] In some embodiments, high-spatial-resolution remote sensing images of the research area can be obtained from public data platforms (such as commercial remote sensing satellite data providers, government agency data platforms, resource satellite application centers, etc.).

[0156] Step 902: Perform zoning control on the research area to obtain flat areas, hillside areas, and forest and grass areas.

[0157] Step 903: Obtain regular cultivated land based on the flat area, terraced fields and sloping cultivated land based on the hillside area, and forest-interspersed cultivated land based on the forest and grass area.

[0158] Step 904: Obtain potential cultivated land plots based on the edge model, regular cultivated land and terraced fields, and obtain potential cultivated land plots based on the texture model, sloping cultivated land and forest-interspersed cultivated land.

[0159] Step 905: Based on the SAR time series data and potential cultivated land plots, perform plot-level time series reconstruction to obtain plot-level time series features.

[0160] Among them, multi-temporal SAR data was used to construct plot-level time series features under boundary constraints.

[0161] Step 906: Obtain the cultivated land plot distribution map based on the time series classification model and the plot-level time series features.

[0162] Among them, based on the classification ability of the recurrent neural network for time series data, combined with the field survey sampling data, the classification abilities of different SAR time series features and their combinations were analyzed and evaluated. Finally, the cultivation and utilization types of each plot were discriminated, so as to eliminate non-cultivated land and further optimize the extraction results.

[0163] Among them, for the crop classification method provided by the related technology, the accuracy of using medium-resolution SAR images to identify crop types is limited. When used to eliminate non-cultivated land, it may misjudge the identification of plots and affect the accuracy of the final result. Using a single deep learning model for cultivated land extraction itself has the problem of missing extraction. Taking this as potential cultivated land, it cannot be supplemented through subsequent processes. Therefore, the phenomenon of missing extraction in the final result is relatively serious. Due to the constraints of the intelligent method itself, it is impossible to correctly extract 100% of the plot information. Some manual correction work is still required to meet the application requirements. The plot distribution map output by the method lacks quality evaluation and cannot guide subsequent manual operations.

[0164] In view of the defects of the above-mentioned related technologies, the present application provides a method for identifying farmland plots based on multi-source data. The similarities with the above-mentioned related technologies include: 1) The goal is to extract cultivated land plots; 2) Deep learning technologies are both adopted; 3) Zoning is used as a prerequisite for implementation during the extraction process; 4) High-resolution and medium-resolution remote sensing image data are both used; The differences from the above-mentioned related technologies include: 1) In the zoning of the solution of the present application, in addition to using the main factors affecting cultivated land types, such as terrain and geomorphic factors, the cultivated land distribution information obtained based on medium-resolution remote sensing images is comprehensively considered, and zoning is carried out according to the characteristics of cultivated land itself. 2) When obtaining the boundary morphology of plots within the zone in the solution of the present application, the integration of different deep learning models such as semantics and edges is comprehensively considered, and the proportion of recognition errors is reduced through type constraints; 3) In the solution of the present application, the medium-resolution remote sensing images are mainly used for pre-identifying the cultivated land scope, rather than for crop type identification, and the accuracy is higher in areas with poor data conditions.

[0165] Figure 10 FIG. is a schematic flowchart of the implementation of a method for identifying farmland plots based on multi-source data provided by an embodiment of the present application. This method can be executed by a processor of an electronic device. This method includes steps S1001 to step S1008, which will be described in combination with Figure 10 the steps shown.

[0166] Step S1001: Preprocess the remote sensing data of the obtained research area.

[0167] In some embodiments, the remote sensing data of the research area includes medium-resolution remote sensing data and high-resolution remote sensing data.

[0168] In some embodiments, preprocessing of the remote sensing data obtained for the study area is performed, including: geometric correction, radiometric correction, cloud detection and removal, position matching between multi-temporal images, and terrain flattening and denoising processing of SAR data, to form a standardized image data product. Among them, the high-resolution image data is synthesized by optimizing the cloud-free data within one year.

[0169] In some embodiments, it is also necessary to classify the cultivated land types in the study area. Among them, the cultivated land in complex scenarios presents various visual characteristics in high-resolution remote sensing images. Such as Figure 11 shown Figure 11 is a schematic diagram of cultivated land types provided by an embodiment of the present application, including: the flat cultivated land 1102 in the flat dam area 1101 is relatively regular, with clear boundaries and obvious textures. The terraced fields 1104 in the mountainous area 1103 have obvious spatial morphological characteristics, with long and narrow plots and clear boundaries; the sloping cultivated land 1105 in the mountainous area 1103 is mainly distributed on the surface of inclined slopes, without clear boundaries and with various spatial morphological characteristics. The forestland cultivated land 1107 in the forest and grass area 1106 has small areas of cultivated land that are scattered and have blurred boundaries in the forest and grass area 1106.

[0170] Step S1002: Calculate the target features of the medium-resolution remote sensing data of the study area.

[0171] In some embodiments, due to complex meteorological conditions, the medium-resolution remote sensing image data obtained at a specific time point cannot cover the entire study area. Therefore, the Sentinel series satellite image data with a spatial resolution of 10-15 meters (including optical satellite and SAR satellite data, 10 meters for the optical satellite and 15 meters for the SAR satellite) is used in the embodiments of the present application. The SAR data is resampled to 10 meters to form a consistent spatial resolution. And calculate features such as spectra, indices, and polarizations of SAR.

[0172] In some embodiments, index features of the target area are calculated based on the spectral features of the remote sensing image data of the optical satellite. Among them, the spectral features include the red, green, blue, near-infrared spectral values, short-wave infrared band 1, and short-wave infrared band 2 of the optical image, and the index features include the normalized difference vegetation index (NVDI), the normalized difference vegetation senescence index (NDSVI), the land surface water index (LSWI), the modified normalized difference water index (MNDWI), the normalized difference yellow index (NDYI), and the normalized difference soil index (NDSI). For the calculation methods of the index features, please refer to Table 1 above.

[0173] In some embodiments, polarization features are calculated based on remote sensing image data of SAR satellites. The polarization modes are divided into horizontal polarization (H) and vertical polarization (V). In the embodiments of the present application, polarization features VH, VV, and cross-polarization ratio (CR) are calculated based on the remote sensing image data of SAR satellites. Among them, when the SAR satellite emits a horizontally polarized microwave pulse and receives a horizontally polarized echo, this mode is denoted as HH polarization; emitting a vertically polarized pulse and receiving a vertically polarized echo is denoted as VV polarization; emitting a horizontally polarized pulse and receiving a vertically polarized echo is HV polarization; emitting a vertically polarized pulse and receiving a horizontally polarized echo is VH polarization. The calculation process refers to the above formula (2).

[0174] In some embodiments, on the basis of the above feature calculation, the present application combines the revisit period of the satellite and the characteristics of the study area, and uses the median synthesis method to calculate the monthly synthesis features of the above spectral features, index features, and SAR features to obtain the target features of the study area.

[0175] Step S1003: Classify the plots in the study area based on the target features of the study area and the random forest model.

[0176] In some embodiments, the random forest model adopted in the present application is trained based on the publicly available European Space Agency (ESA) series datasets and the China Land Cover Dataset (CLCD).

[0177] In some embodiments, the obtained target features and the medium-resolution remote sensing data of the study area are input into the trained random forest model to obtain the plot classification according to five land use types: cultivated land, forest and grassland, built-up area, water body area, and others.

[0178] Step S1004: Obtain the regular cultivated land in the study area.

[0179] In some embodiments, the extraction range of regular cultivated land needs to be obtained first, including: obtaining the elevation data of the research area, dividing the research area based on the elevation data with a 12° boundary, where the flat dam area has a slope less than 12° and the mountainous area has a slope greater than 12°. Secondly, overlay the above-obtained 5 types of land parcel classifications on the flat dam area distribution map, and remove the built-up area and water area within the flat dam area to obtain the extraction range of regular cultivated land. Secondly, obtain the regular cultivated land in the research area from the extraction range of regular cultivated land, including: overlay the extraction range of regular cultivated land with the road and water system vector layers matched with the high-resolution image of the extraction range of regular cultivated land, and further divide the extraction range of regular cultivated land into multiple extraction task blocks; for each task extraction block, use the Dense Extreme Inception Network for Edge Detection (DexiNed) model to perform edge detection on the high-resolution remote sensing image of each task extraction block to obtain an edge intensity map, and then use the vectorization method to obtain the distribution map of suspected regular cultivated land patches within the block. Then, use the U-net++ semantic segmentation model to segment the high-resolution remote sensing image of each task extraction block to obtain the segmentation result indicating whether the task extraction block is regular cultivated land. Finally, overlay the distribution map of suspected regular cultivated land patches with the segmentation result, and take the intersection of the land patches as the regular cultivated land within each block.

[0180] Step S1005: Obtain the forested cultivated land in the research area.

[0181] In some embodiments, for the forest and grassland in the research area divided into 5 types of land parcels, calculate the proportion of the area of the forest and grassland type within the grid to the total area of the grid according to a 100-meter grid. If the area of the forest and grass type within the grid accounts for more than 80% of the total area of the grid, it is used as the extraction range of forested cultivated land. The calculation method can refer to the above formula (2). Overlay the extraction range of forested cultivated land with the road, water system, ridge line, and gully line vector layers matched with the high-resolution image of the extraction range of forested cultivated land, and divide the extraction range of forested cultivated land into multiple task extraction blocks. For each task extraction block, based on the high-resolution remote sensing image of each task extraction block, use the Deeplab v3+ model for segmentation to obtain the segmentation result indicating whether the task extraction block is forested cultivated land. Overlay the segmentation result with the cultivated land in the research area divided into 5 types of land parcels, retain the segmentation patches within the cultivated land range, and for those without patches extracted within the cultivated land parcel range, generate patches based on the raster-to-vector conversion of the classification result map. The two are integrated to form the forested cultivated land parcels within the task block.

[0182] Step S1006: Obtain the terraced cultivated land in the research area.

[0183] In some embodiments, according to the standard that terraced fields can be built in mountainous areas with a slope below 25 degrees, the mountainous areas in the study area are divided into two levels with 25° as the boundary. The areas with a slope less than or equal to 25° are terraced field extraction areas. The cultivated land in the study area divided into 5 types of plots is overlaid on the distribution map of the terraced field extraction areas, and the built-up areas, water areas and forest and grassland areas within the scope are excluded to obtain the extraction scope of the terraced fields. The extraction scope of the terraced fields is overlaid with the road and water system vector layers matched with the high-resolution images, and the extraction scope of the terraced fields is further divided into multiple extraction task blocks. For each task extraction block, based on the high-resolution remote sensing image of the block, the RCF edge detection model is used to extract the fine boundary to obtain the edge intensity map, and then the vectorization method is used to obtain the distribution map of suspected terraced field patches within the task block. Then, the semantic segmentation model U-net++ model is selected for segmentation to generate the segmentation result indicating whether it is a terraced field. Finally, the distribution map of suspected terraced field patches is overlaid with the segmentation result, and the plot patches of the intersection are used as the distribution map of the terraced fields within the task block.

[0184] Step S1007: Obtain the sloping cultivated land in the study area.

[0185] In some embodiments, considering that sloping cultivated land is widely distributed in mountainous areas (areas with a slope greater than 12 degrees), the area between 12 degrees and 25 degrees belongs to the area where both terraced fields and sloping cultivated land exist. The mountainous area range divided by the study area based on a slope greater than 12° is overlaid with the cultivated land range in the study area divided into 5 types of plots, and the built-up areas, water areas and forest and grassland areas within the mountainous area range are excluded to obtain the extraction scope of the sloping cultivated land.

[0186] The extraction scope of the sloping cultivated land is overlaid with the road and water system vector layers matched with the high-resolution images to divide the extraction scope of the sloping cultivated land into multiple extraction task blocks. For each task extraction block, the U-net model is used for segmentation based on the high-resolution remote sensing image of the task extraction block to obtain the distribution map of suspected sloping cultivated land.

[0187] Step S1008: Integrate the regular cultivated land, forest-interspersed cultivated land, terraced fields and sloping cultivated land in the above-mentioned study area to obtain the cultivated land distribution in the study area.

[0188] In some embodiments, the distribution map of different types of cultivated land is used as a spatial constraint, combined with the recognition accuracy of different types of cultivated land, and in the same partition, the results of extraction from different models are integrated in the order of result accuracy from high to low. When there is a conflict in the extraction results of multiple models, the model result with high accuracy is selected as the final result map. It can be understood that for the flat area where the study area is divided based on a slope of less than 12°, the range of forest and grassland in the study area divided into 5 types of plots is used as a mask, and the type of cultivated land outside the mask area is determined as the plot distribution of regular cultivated land, and the type of cultivated land within the mask area is determined as the plot distribution of forest cultivated land; for the mountainous area where the study area is divided based on a slope greater than 12°, first, the mountainous area above 25° is used as a mask with the range of forest and grassland in the study area divided into 5 types of plots, and the type of cultivated land outside the mask area is determined as the plot distribution of sloping cultivated land, and the type of cultivated land within the mask area is determined as the plot distribution of forest cultivated land; for the mountainous area with a slope of 12° For areas with an angle of up to 25°, the forest and grassland range in the study area divided into five types of plots was also used as a mask, and the plot distribution within the mask area was determined as the forest cultivated land; for areas outside the mask area, the non-repeated parts of the terrace extraction results and the sloping cultivated land extraction results were respectively determined as the plot distribution of terraces and the plot distribution of sloping cultivated land, and for the repeated parts of the terrace extraction results and the sloping cultivated land extraction results, the terraces were used as the repeated part of the plot distribution, and the above-mentioned plot distribution of regular cultivated land, plot distribution of forest cultivated land, plot distribution of terraces, and plot distribution of sloping cultivated land were synthesized to obtain the plot distribution of different cultivated land types in the study area.

[0189] In the embodiment of the present application, the cultivated land range is first identified based on the medium-resolution image data, which can effectively reduce the problem of missed cultivated land identification due to incomplete high-resolution data. At the same time, combined with data such as elevation, it can effectively divide the distribution areas of different types of cultivated land such as regular cultivated land, exclude non-cultivated areas, and reduce the workload of subsequent extraction. Through the classification of cultivated land types, layered extraction can select more suitable methods for different types, thereby improving the overall accuracy of identification. Using partitions as integrated constraints can reduce the post-processing work caused by subsequent spatial conflicts.

[0190] Based on the foregoing embodiments, an embodiment of the present application provides a farmland plot recognition device based on multi-source data. The device includes each unit included therein, as well as each module included in each unit, and can be implemented by a processor in an electronic device; of course, it can also be implemented by a specific logic circuit; during implementation, the processor can be a central processing unit (CPU), a microprocessor unit (MPU), a digital signal processor (DSP), or a field programmable gate array (FPGA), etc.

[0191] Figure 12 FIG. is a schematic structural diagram of a farmland plot recognition device based on multi-source data provided by an embodiment of the present application. As Figure 12 shown, the farmland plot recognition device 1200 based on multi-source data includes: an acquisition module 1201, a division module 1202, and a generation module 1203, where: the acquisition module 1201 is configured to acquire first remote sensing data and second remote sensing data of a research area; wherein, the resolution of the second remote sensing data is higher than that of the first remote sensing data; the division module 1202 is configured to divide the research area into at least one type of target area based on the first remote sensing data and a first target model; the at least one type of target area represents the plot distribution of different land use types in the research area; the generation module 1203 is configured to generate a target category sub-block corresponding to the research area based on the at least one type of target area, the second remote sensing data, and a second target model; the target category sub-block represents the plot distribution of the target land use type in the research area.

[0192] In some embodiments, the generation module 1203 is further configured to generate at least one type of target plot corresponding to the research area based on the at least one type of target area, the second remote sensing data, and a second target model; and generate a target category sub-block corresponding to the research area based on the at least one type of target plot.

[0193] In some embodiments, the at least one type of target plot includes regular cultivated land plots; the second target model includes a first semantic segmentation model; the generating module 1203 is further configured to obtain a first target object for cultivated land area segmentation in a first area based on second remote sensing data; the first area includes an area in the study area with a slope less than a first preset slope; segment the first area based on the first target object to obtain at least one first sub-area; obtain first suspected regular cultivated land plots in at least one first sub-area through a vectorization method based on the second remote sensing data; obtain second suspected regular cultivated land plots in at least one first sub-area based on at least one first sub-area, the first semantic segmentation model, and the second remote sensing data; superimpose the first suspected regular cultivated land plots and the second suspected regular cultivated land plots in at least one first sub-area to generate the regular cultivated land plots.

[0194] In some embodiments, the at least one type of target plot includes forest-interspersed cultivated land plots; the at least one type of target area includes forest-grass areas and cultivated land areas; the second target model includes a second semantic segmentation model; the generating module 1203 is further configured to obtain a second target object for cultivated land area segmentation in a target forest-grass plot; the target forest-grass area represents an area in the forest-grass area where the forest-grass land area reaches a preset threshold; segment the target forest-grass area based on the second target object to obtain at least one second sub-area; generate suspected forest-interspersed cultivated land plots in at least one second sub-area based on at least one second sub-area, the second remote sensing data, and the second semantic segmentation model; superimpose the suspected forest-interspersed cultivated land plots and the cultivated land areas to generate the forest-interspersed cultivated land plots.

[0195] In some embodiments, the at least one type of target plot includes terraced cultivated land plots; the second target model includes a third semantic segmentation model; the generating module 1203 is further configured to screen a second area in the study area to obtain a first screening area; the second area includes an area in the study area with a slope greater than the first preset slope; the first screening area includes an area in the second area with a slope greater than the first preset slope and less than or equal to a second preset slope; obtain a third target object for cultivated land area segmentation in the first screening area; segment the first screening area based on the third target object to obtain at least one third sub-area; obtain first suspected terraced cultivated land plots in at least one third sub-area through a vectorization method based on the second remote sensing data; obtain second suspected terraced cultivated land plots in at least one third sub-area based on at least one third sub-area, the second remote sensing data, and the third semantic segmentation model; superimpose the first suspected terraced cultivated land plots and the second suspected terraced cultivated land plots in at least one third sub-area to generate the terraced cultivated land plots.

[0196] In some embodiments, the at least one type of target plot includes sloping cultivated land plots; the second target model includes a fourth semantic segmentation model; the generating module 1203 is further configured to screen a second area in the study area to obtain a second screening area; the second area includes an area in the study area with a slope greater than a first preset slope; the first screening area includes an area in the second area with a slope greater than a second preset slope; wherein the second preset slope is greater than the first preset slope; obtain a fourth target object for cultivated land area segmentation in the second screening area; segment the second screening area based on the fourth target object to obtain at least one fourth sub-area; and generate the sloping cultivated land plots based on the at least one fourth sub-area, the second remote sensing data, and the fourth semantic segmentation model.

[0197] In some embodiments, the partitioning module 1202 is further configured to generate target feature data of the study area based on the first remote sensing data; and generate at least one target area of the study area based on the target feature data and a first target model.

[0198] In some embodiments, the first remote sensing data includes monthly optical data and monthly SAR data; the target feature data includes target index features and target SAR features; the partitioning module 1202 is further configured to generate monthly index features based on the monthly optical data, and generate monthly SAR features based on the monthly SAR data; synthesize the monthly index features to obtain target index features, and synthesize the monthly SAR features to obtain target SAR features.

[0199] The description of the above device embodiments is similar to the description of the above method embodiments and has similar beneficial effects to the method embodiments. In some embodiments, the functions or modules included in the device provided in the embodiments of the present application can be used to execute the methods described in the above method embodiments. For technical details not disclosed in the device embodiments of the present application, please refer to the description of the method embodiments of the present application for understanding.

[0200] It should be noted that in the embodiments of the present application, if the above-mentioned method is implemented in the form of software function modules and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of the present application, in essence or the part that contributes to the related technology, can be embodied in the form of a software product. The software product is stored in a storage medium and includes several instructions for causing an electronic device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in the embodiments of the present application. The aforementioned storage medium includes: various media that can store program codes such as USB flash drives, mobile hard disks, read-only memories (ROMs), magnetic disks, or optical discs. In this way, the embodiments of the present application are not limited to any specific hardware, software, or firmware, or any combination among hardware, software, and firmware.

[0201] The embodiments of the present application provide an electronic device, including a memory and a processor. The memory stores a computer program that can run on the processor, and when the processor executes the program, it implements some or all of the steps in the above-mentioned method.

[0202] The embodiments of the present application provide a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements some or all of the steps in the above-mentioned method. The computer-readable storage medium can be transient or non-transient.

[0203] The embodiments of the present application provide a computer program, including computer-readable code. When the computer-readable code runs in an electronic device, the processor in the electronic device executes to implement some or all of the steps in the above-mentioned method.

[0204] The embodiments of the present application provide a computer program product. The computer program product includes a non-transient computer-readable storage medium storing a computer program. When the computer program is read and executed by a computer, it implements some or all of the steps in the above-mentioned method. The computer program product can be specifically implemented in the form of hardware, software, or a combination thereof. In some embodiments, the computer program product is specifically embodied as a computer storage medium. In other embodiments, the computer program product is specifically embodied as a software product, such as a software development kit (SDK), etc.

[0205] It should be noted here that the descriptions of the above embodiments tend to emphasize the differences between the embodiments, and their similarities can be referred to each other. The descriptions of the above embodiments of the device, storage medium, computer program, and computer program product are similar to the descriptions of the above method embodiments and have beneficial effects similar to those of the method embodiments. For the technical details not disclosed in the embodiments of the device, storage medium, computer program, and computer program product of the present application, please refer to the descriptions of the method embodiments of the present application for understanding.

[0206] Figure 13 The following is a schematic diagram of the hardware entity of an electronic device provided by an embodiment of the present application. As Figure 13 shown, the hardware entity of the electronic device 1300 includes: a processor 1301 and a memory 1302. Among them, the memory 1302 stores a computer program that can run on the processor 1301, and when the processor 1301 executes the program, it implements the steps in the method of any of the above embodiments.

[0207] The memory 1302 stores a computer program that can run on the processor. The memory 1302 is configured to store instructions and applications executable by the processor 1301, and can also cache data to be processed or already processed by the processor 1301 and each module in the electronic device 1300 (for example, image data, audio data, voice communication data, and video communication data), and can be implemented by flash memory (FLASH) or random access memory (Random Access Memory, RAM).

[0208] When the processor 1301 executes the program, it implements the steps of the method of any of the above items. The processor 1301 generally controls the overall operation of the electronic device 1300.

[0209] An embodiment of the present application provides a computer storage medium. The computer storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of the method of any of the above embodiments.

[0210] It should be noted here that the descriptions of the above storage medium and device embodiments are similar to the descriptions of the above method embodiments and have beneficial effects similar to those of the method embodiments. For the technical details not disclosed in the embodiments of the storage medium and device of the present application, please refer to the descriptions of the method embodiments of the present application for understanding.

[0211] The above-mentioned processor may be at least one of an Application Specific Integrated Circuit (ASIC), a Digital Signal Processor (DSP), a Digital Signal Processing Device (DSPD), a Programmable Logic Device (PLD), a Field Programmable Gate Array (FPGA), a Central Processing Unit (CPU), a controller, a microcontroller, and a microprocessor. It can be understood that the electronic device implementing the functions of the above-mentioned processor may also be others, which are not specifically limited in the embodiments of the present application.

[0212] The above-mentioned computer storage medium / memory may be a Read Only Memory (ROM), a Programmable Read-Only Memory (PROM), an Erasable Programmable Read-Only Memory (EPROM), an Electrically Erasable Programmable Read-Only Memory (EEPROM), a Ferromagnetic Random Access Memory (FRAM), a Flash Memory, a magnetic surface memory, an optical disc, or a Compact Disc Read-Only Memory (CD-ROM), etc.; or it may also be various terminals including one or any combination of the above-mentioned memories, such as a mobile phone, a computer, a tablet device, a personal digital assistant, etc.

[0213] It should be understood that the "one embodiment" or "an embodiment" mentioned throughout the specification means that the specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, the "in one embodiment" or "in an embodiment" that appears throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in various embodiments of the present application, the magnitude of the serial numbers of the above steps / processes does not mean the order of execution, and the execution order of each step / process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application. The serial numbers of the embodiments of the present application above are only for description and do not represent the advantages or disadvantages of the embodiments.

[0214] It should be noted that in this article, the term "comprising", "including" or any other variant thereof is intended to cover a non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the phrase "comprising a..." does not exclude the existence of additional identical elements in the process, method, article or device including such element.

[0215] In several embodiments provided by the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only illustrative. For example, the division of the units is only a logical function division, and there can be other division methods in actual implementation. For example, multiple units or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the coupling, direct coupling or communication connection between the components shown or discussed with each other can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0216] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units; they can be located in one place or distributed to multiple network units; some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0217] In addition, in each embodiment of the present application, each functional unit can be entirely integrated into one processing unit, or each unit can be separately regarded as one unit, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware, or in the form of a combination of hardware and software functional units. Those of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps included in the above method embodiments. The aforementioned storage medium includes: removable storage devices, read-only memory (ROM), magnetic disks, or optical discs and other various media that can store program codes.

[0218] Alternatively, if the above integrated unit of the present application is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence or the part that contributes to the related art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing an electronic device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes: removable storage devices, ROM, magnetic disks, or optical discs and other various media that can store program codes.

[0219] As described above, only the embodiments of the present application are provided, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application.

Claims

1. A method for identifying farmland plots based on multi-source data, characterized in that: The method comprises: Acquire first remote sensing data and second remote sensing data of a study area; wherein the resolution of the second remote sensing data is higher than the resolution of the first remote sensing data; Based on the first remote sensing data and the first target model, the study area is divided into at least one type of target area; the at least one type of target area represents the distribution of plots of different land use types in the study area; Based on the at least one type of target area, the second remote sensing data and the second target model, a target category block corresponding to the study area is generated; the target category block represents the distribution of plots of target land use type in the study area.

2. The method according to claim 1, characterized in that The generating target category blocks corresponding to the target area based on the at least one type of target area, the second remote sensing data and the second target model includes: Based on the at least one type of target area, the second remote sensing data and the second target model, generating at least one type of target plot corresponding to the study area; Generate target category blocks corresponding to the study area based on the at least one type of target plot.

3. The method according to claim 2, characterized in that The at least one type of target plot includes a regular cultivated land plot; the second target model includes a first semantic segmentation model; The generating, based on the at least one type of target area, the second remote sensing data and the semantic segmentation model, at least one type of target plot corresponding to the study area comprises: Based on the second remote sensing data, a first target object for cultivated land area segmentation in a first area is obtained; the first area includes an area in the study area whose slope is less than a first preset slope; Partition the first area based on the first target object to obtain at least one first sub-area; Based on the second remote sensing data, obtaining a first suspected regular cultivated land parcel in at least one first sub-region by a vectorization method; Based on at least one first sub-region, the first semantic model and the second remote sensing data, obtaining a second suspected regular cultivated land plot in at least one first sub-region; The first suspected regular farmland block and the second suspected regular farmland block in the at least one first sub-region are superimposed to generate the regular farmland block.

4. The method according to claim 2, characterized in that: The at least one type of target plot includes forested farmland plots; the at least one type of target area includes forest and grass areas and farmland areas; the second target model includes a second semantic segmentation model; The generating at least one type of target plot corresponding to the study area based on the at least one type of target area, the second remote sensing data and the second target model includes: Acquire a second target object for segmenting the cultivated land area in the target forest and grassland area; the target forest and grassland area represents an area in the forest and grassland area where the forest and grassland area reaches a preset threshold; Segmenting the target forest and grassland area based on the second target object to obtain at least one second sub-area; Based on at least one second sub-region, the second remote sensing data and the second semantic segmentation model, generating a suspected forest-cultivated land parcel in the at least one second sub-region; The suspected forest-based cultivated land plot and the cultivated land area are superimposed to generate the forest-based cultivated land plot.

5. The method according to claim 2, characterized in that: The at least one type of target plot includes terraced farmland plots; the second target model includes a third semantic segmentation model; The generating at least one type of target plot corresponding to the study area based on the at least one type of target area, the second remote sensing data and the second target model includes: Screening the second area in the study area to obtain a first screening area; the second area includes an area in the study area whose slope is greater than a first preset slope; the first screening area includes an area in the second area whose slope is greater than the first preset slope and less than or equal to a second preset slope; Acquire a third target object for farmland area segmentation in the first screening area; Segmenting the first screening area based on the third target object to obtain at least one third sub-area; Based on the second remote sensing data, obtaining a first suspected terraced farmland plot in the at least one third sub-region by a vectorization method; Based on the at least one third sub-region, the second remote sensing data and the third semantic segmentation model, obtaining a second suspected terraced farmland plot in the at least one third sub-region; The first suspected terraced farmland plot and the second suspected terraced farmland plot in the at least one third sub-region are superimposed to generate the terraced farmland plot.

6. The method according to claim 2, characterized in that The at least one type of target plot includes sloping cultivated land; the second target model includes a fourth semantic segmentation model; The generating at least one type of target plot corresponding to the study area based on the at least one type of target area, the second remote sensing data and the second target model includes: Screening a second area in the study area to obtain a second screening area; the second area includes an area in the study area with a slope greater than a first preset slope; the second screening area includes an area in the second area with a slope greater than a second preset slope; wherein the second preset slope is greater than the first preset slope; Acquire a fourth target object for farmland area segmentation in the second screening area; Segmenting the second screening area based on the fourth target object to obtain at least one fourth sub-area; The sloping farmland plot is generated based on the at least one fourth sub-region, the second remote sensing data and the fourth semantic segmentation model.

7. The method according to any one of claims 1 to 6, characterized in that: The dividing the first remote sensing data into at least one type of target area based on the first remote sensing data and the first target model includes: generating target feature data of the study area based on the first remote sensing data; At least one target region of the study area is generated based on the target feature data and the first target model.

8. The method according to claim 7, characterized in that The first remote sensing data includes monthly optical data and monthly synthetic aperture radar SAR data; the target feature data includes target index features and target SAR features; The step of generating target feature data of the research area based on the first remote sensing data comprises: generating a monthly index feature based on the monthly optical data, and generating a monthly SAR feature based on the monthly SAR data; The monthly index features are synthesized to obtain a target index feature, and the monthly SAR features are synthesized to obtain a target SAR feature.

9. A farmland plot identification device based on multi-source data, characterized in that: The device comprises: An acquisition module, used for acquiring first remote sensing data and second remote sensing data of a research area; wherein the resolution of the second remote sensing data is higher than the resolution of the first remote sensing data; A division module, configured to divide the study area into at least one type of target area based on the first remote sensing data and the first target model; the at least one type of target area represents the distribution of plots of different land use types in the study area; A generation module is used to generate target category blocks corresponding to the study area based on the at least one type of target area, the second remote sensing data and the second target model; the target category blocks represent the plot distribution of the target land use type in the study area.

10. An electronic device comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, characterized in that: When the processor executes the program, the steps in the method according to any one of claims 1 to 7 are implemented.

Images