Planting plot space structure reconstruction method, system and device and storage medium

Abstract

The invention discloses a planting plot space structure reconstruction method, system and device and a storage medium, and relates to the technical field of remote sensing application. The method comprises the following steps: acquiring a remote sensing image of a target planting area; performing partition deconstruction according to the remote sensing image to obtain a plurality of farming areas and corresponding area types; the region types comprise paddy fields, regular planting plots, fuzzy planting plots, terraced fields and garden plots; selecting a corresponding image extraction model according to the region type, and predicting the image of each farming region by using the image extraction model to obtain a predicted intensity map; the prediction intensity map comprises an edge prediction intensity map and a surface prediction intensity map; the image extraction model comprises a conventional edge model, a fine edge model and a texture model; performing grid vectorization processing on the prediction intensity map to obtain a pattern spot vector surface; and performing vector fusion processing on each pattern spot vector surface to obtain plot pattern spot data. According to the invention, the extraction precision of the planting plot can be improved.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing application technology, and in particular to a method, system, device and storage medium for reconstructing the spatial structure of a planting plot. Background Art

[0002] China's mountainous areas account for a large proportion of agriculture, and have complex terrain conditions, large slopes of cultivated land, small areas, and fragmented distribution, making it difficult to quickly and accurately extract the vector boundaries of mountain plots. The traditional "fine plot" production method is generally to compare remote sensing images in GIS software, form a clear visual distinction with surrounding objects through significant texture mutations or linear boundaries, and clearly outline along the texture mutations or linear boundaries to depict the plots in the form of surface elements, which is time-consuming and laborious. With the development of artificial intelligence technology, automated information extraction has been achieved, but in many existing agricultural planting plot extraction technology studies, there is no starting point from the planting plots themselves, and the existing models are directly used for brute force, resulting in low precision of automated extraction products of mountain plots, which is far from "fine plots". The diversity of cultivated land types in mountainous areas cannot be analyzed by a single model (heterogeneity). Therefore, how to quickly and accurately extract "fine plots" through high-resolution images has become a difficult problem. Summary of the invention

[0003] The purpose of the present invention is to provide a method, system, device and storage medium for reconstructing the spatial structure of a planting plot, which can improve the extraction accuracy of the planting plot.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] A method for reconstructing the spatial structure of a planting plot, comprising:

[0006] Obtain remote sensing images of the target planting area;

[0007] Deconstructing the remote sensing images by partitioning to obtain a plurality of farming areas and corresponding area types; the area types include paddy fields, regular planting plots, fuzzy planting plots, terraces and gardens;

[0008] Selecting a corresponding image extraction model according to the type of region, and using the image extraction model to predict the image of each farming area to obtain a prediction intensity map; the prediction intensity map includes an edge prediction intensity map and a surface prediction intensity map; the image extraction model includes a conventional edge model, a fine edge model and a texture model; the conventional edge model and the texture model are constructed based on the D-LinkNet network model; the fine edge model is constructed based on the RCF edge detector;

[0009] Performing raster vectorization processing on the predicted intensity map to obtain a spot vector surface;

[0010] Vector fusion processing is performed on each of the above-mentioned map spot vector surfaces to obtain land plot map spot data.

[0011] Optionally, the zoning and deconstructing according to the remote sensing image to obtain a plurality of farming areas and corresponding area types specifically includes:

[0012] Collect administrative data and DEM raster images of the target planting area;

[0013] Using the remote sensing image as a base map and performing zoning of “administrative district-planting district-cultivation district-plot” according to the administrative data, the boundaries of administrative districts are determined;

[0014] Calculate regional terrain data based on the DEM raster image and draw the boundary of the planting area;

[0015] Merging the administrative division boundary and the planting area boundary to obtain a division control network;

[0016] Image clipping is performed based on the regional terrain data and the zoning control network to obtain multiple farming areas and corresponding area types.

[0017] Optionally, before the remote sensing image is used as the base map and the administrative data is used to divide the “administrative area-planting area-cultivation area-plot” into different areas and the administrative area boundaries are determined, the method further includes:

[0018] The remote sensing image is subjected to image preprocessing; the image preprocessing includes orthorectification, radiation calibration, atmospheric correction, cloud removal, geometric correction, image enhancement and image mosaicking.

[0019] Optionally, the selecting a corresponding image extraction model according to the region type, and using the image extraction model to predict the image of each of the farming areas to obtain an edge prediction intensity map specifically includes:

[0020] When the area type is a regular planting plot, the conventional edge model and the texture model are used in turn to predict the edge of the area to obtain an edge prediction intensity map;

[0021] When the area type is terraced fields, the edge of the area is predicted using a fine edge model and a texture model in turn to obtain an edge prediction intensity map;

[0022] When the area type is fuzzy planting plots, gardens and paddy fields, the texture model and the conventional edge model are used in turn to perform predictions to obtain an edge prediction intensity map.

[0023] Optionally, performing raster vectorization processing on the predicted intensity map to obtain a spot vector surface specifically includes:

[0024] Extracting skeleton lines of the edge prediction strength map, connecting broken lines and then vectorizing them to obtain vector lines;

[0025] Simplifying and straightening the vector line and then converting it into a first vector surface;

[0026] After dilation, hole filling and binarization processing are performed on the surface prediction strength map, a second vector surface is obtained;

[0027] A pattern vector plane is determined according to the first vector plane and the second vector plane.

[0028] The present invention also provides a planting plot spatial structure reconstruction system, comprising:

[0029] Remote sensing data acquisition module, used to obtain remote sensing images of the target planting area;

[0030] A partition structure module is used to partition and deconstruct the remote sensing image to obtain multiple farming areas and corresponding area types; the area types include paddy fields, regular planting plots, fuzzy planting plots, terraces and gardens;

[0031] An edge prediction module is used to select a corresponding image extraction model according to the type of the region, and use the image extraction model to predict the image of each of the farming areas to obtain a prediction intensity map; the prediction intensity map includes an edge prediction intensity map and a surface prediction intensity map; the image extraction model includes a conventional edge model, a fine edge model and a texture model; the conventional edge model and the texture model are constructed based on the D-LinkNet network model; the fine edge model is constructed based on the RCF edge detector;

[0032] A vector processing module, used for performing raster vectorization processing on the predicted intensity map to obtain a spot vector surface;

[0033] The vector fusion module is used to perform vector fusion processing on each of the map spot vector surfaces to obtain the plot map spot data.

[0034] The present invention also provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the above-mentioned method for reconstructing the spatial structure of a planting plot.

[0035] The present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the method for reconstructing the spatial structure of a planting plot as described above.

[0036] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0037] The present invention discloses a method, system, device and storage medium for reconstructing the spatial structure of a planting plot. The method comprises the following steps: obtaining a remote sensing image of a target planting area; performing partition deconstruction according to the remote sensing image to obtain a plurality of farming areas and corresponding area types; the area types include paddy fields, regular planting plots, fuzzy planting plots, terraces and gardens; selecting a corresponding image extraction model according to the area type, and using the image extraction model to predict the image of each farming area to obtain a prediction intensity map; the prediction intensity map includes an edge prediction intensity map and a surface prediction intensity map; the image extraction model includes a conventional edge model, a fine edge model and a texture model; performing raster vectorization processing on the prediction intensity map to obtain a patch vector surface; performing vector fusion processing on each patch vector surface to obtain plot patch data.

[0038] The present invention uses the partitioning idea to partition the data, which can simplify the complex surface and concentrate the elements of similar plot characteristics into several relatively independent sub-areas, reducing the difficulty of extracting objects. After partitioning, the image with a large amount of data can be divided into several small areas, which can effectively disperse the data volume of a single image and reduce the possibility of problems caused by the large amount of data during the extraction process. At the same time, multi-machine multi-task parallel processing can also be performed, which can significantly shorten the extraction time and improve the extraction accuracy of the planting plots. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0040] Figure 1 It is a schematic diagram of the flow chart of the method for reconstructing the spatial structure of a planting plot of the present invention;

[0041] Figure 2 This is a logical diagram of the extraction process in this embodiment;

[0042] Figure 3 This is a schematic diagram of the deep learning extraction process in this embodiment. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] The purpose of the present invention is to provide a method, system, device and storage medium for reconstructing the spatial structure of a planting plot, which can improve the extraction accuracy of the planting plot.

[0045] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0046] like Figure 1-Figure 3 As shown, the present invention provides a method for reconstructing the spatial structure of a planting plot, comprising:

[0047] Step 100: Acquire a remote sensing image of the target planting area;

[0048] Step 200: performing partition deconstruction according to the remote sensing image to obtain multiple farming areas and corresponding area types; the area types include paddy fields, regular planting plots, fuzzy planting plots, terraces and gardens; specifically including:

[0049] Collect administrative data and DEM raster images of the target planting area; use the remote sensing image as a base map, and divide the "administrative area-planting area-cultivation area-plot" into zones according to the administrative data to determine the administrative zone boundaries; calculate regional terrain data according to the DEM raster image, and draw the planting area boundaries; merge the administrative zone boundaries and the planting area boundaries to obtain a zone control network; perform image cropping based on the regional terrain data and the zone control network to obtain multiple cultivation areas and corresponding area types.

[0050] The remote sensing images are further subjected to image preprocessing before application; the image preprocessing includes orthorectification, radiation calibration, atmospheric correction, cloud removal, geometric correction, image enhancement and image mosaicking.

[0051] Step 300: Selecting a corresponding image extraction model according to the region type, and using the image extraction model to predict the image of each farming area to obtain a prediction intensity map; the prediction intensity map includes an edge prediction intensity map and a surface prediction intensity map; the image extraction model includes a conventional edge model, a fine edge model and a texture model; the conventional edge model and the texture model are constructed based on the D-LinkNet network model; the fine edge model is constructed based on the RCF edge detector;

[0052] In specific applications:

[0053] When the area type is a regular planting plot, the conventional edge model and the texture model are used in turn to predict the edge of the area to obtain an edge prediction intensity map; when the area type is a terraced field, the fine edge model and the texture model are used in turn to predict the edge of the area to obtain an edge prediction intensity map; when the area type is a fuzzy planting plot, a garden, and a paddy field, the texture model and the conventional edge model are used in turn to predict to obtain an edge prediction intensity map.

[0054] Step 400: performing raster vectorization processing on the predicted intensity map to obtain a spot vector surface; specifically comprising:

[0055] Extract the skeleton line of the edge prediction strength map, connect the broken lines and then vectorize them to obtain vector lines; simplify and straighten the vector lines and then convert them into a first vector surface; expand, fill holes and binarize the surface prediction strength map to obtain a second vector surface; determine the pattern vector surface based on the first vector surface and the second vector surface.

[0056] Step 500: Perform vector fusion processing on each of the above-mentioned spot vector surfaces to obtain land spot data.

[0057] Based on the above technical solution, the following embodiments are provided.

[0058] In this embodiment, according to the current "fine plot" map extraction needs, high-resolution remote sensing images are used as basic data, and based on administrative boundaries, elevations, slopes, slope directions, plots and other factors, combined with the idea of ​​geographical zoning, the complex surface is deconstructed, and multiple independent sub-areas (cultivation areas) of similar scales are finally formed according to the zoning structure of administrative area-planting area-cultivation area-plot. Each cultivation area has the same internal terrain, similar crop types, similar plot textures, and uniform overall image features, which ultimately achieves large-scale, high-precision, and high-efficiency mountain plot extraction.

[0059] Extraction steps: According to the texture characteristics of the plots in each farming area and their visual characteristics on the image, the corresponding extraction method is selected to extract the spots, and finally the plot spot data is obtained by layer updating.

[0060] For the zoning structure of "administrative district-planting district-cultivation district-plot":

[0061] Administrative Region:

[0062] The government generally encourages farmers to grow food crops through subsidy policies, and also promotes farmers to grow high-value-added specialty cash crops through guiding policies. Different agricultural policies will directly or indirectly affect the types of crops that farmers choose to plant, and the image texture features of plots of the same crop type are more similar. Therefore, based on high-resolution remote sensing images, with township-level administrative vector boundaries as reference, the study area is visually partitioned, and then the extraction error caused by element differences is reduced, and the initial classification of plots is achieved to improve the extraction accuracy.

[0063] Zoning steps: ① Ensure that the integrity of the plot is not destroyed; ② Refer to the administrative vector boundary for zoning. Finally, the administrative zoning surface feature shp file is formed.

[0064] Planting area:

[0065] Planting zoning refers to the secondary zoning of the study area based on administrative zoning, according to natural geographical features such as topography and landforms, and the division of different landforms such as mountains, hills and plains into different planting zones. Determine the characteristics of each planting area, including crop type, priority order (weight), phenological characteristics, etc., to reduce the differences in crop types and phenological characteristics within each planting area. The zoning standards are shown in Table 1.

[0066] Zoning steps: ① Ensure that the integrity of the plot is not destroyed; ② By calculating DEM data, different landforms such as mountains, hills and plains are divided into different planting zones. Finally, the planting zone surface element shp file is formed.

[0067] Table 1 Planting zoning standards

[0068]

[0069]

[0070] Cultivated area:

[0071] Cultivated area refers to the division of the cultivated land into paddy fields, regular cultivated land, fuzzy cultivated land, terraced fields and gardens based on the morphology and texture characteristics of the cultivated land in the cultivated land. The zoning standards are shown in Table 2 and the cultivation zoning standards are shown in Table 3. The image characteristics of cultivated land plots in each cultivated area determine the plot extraction model and fusion order.

[0072] Zoning steps: ① Ensure that the integrity of the plot is not destroyed; ② Through visual interpretation, the different cultivated land types corresponding to paddy fields, regular planting plots, fuzzy planting plots, terraces, and gardens are divided into different zones. Finally, the cultivated area surface element shp file is formed.

[0073] Table 2 Planting zones correspond to farming zones

[0074] Planting Zone Corresponding farming zones Plateau and flat areas Regular planting plots, gardens Pingbaping area Paddy fields, regular planting plots, gardens Hilly slope area Paddy fields, terraces, gardens Mountain slope area Terraces and gardens Forest and grassland area Fuzzy planting plots

[0075] Table 3 Cultivation zoning standards

[0076]

[0077]

[0078] Then, “fine plot” stratified extraction is performed.

[0079] According to the texture characteristics of the plots in each farming area, appropriate extraction schemes are selected for extraction. The extraction steps follow the basic steps of deep learning: "sample preparation-model training-image prediction-raster vectorization-vector post-processing".

[0080] 1) First, conduct an overall analysis of the entire plot in each partition, select cultivated land samples evenly within the image range, and draw the cultivated land boundary information. 2) Convert the sample drawing results into binary results as label data and training data for the edge monitoring network. By training boundary features, obtain the extraction model of cultivated land boundary information. 3) Predict the cropped images of each task partition to obtain the boundary prediction intensity result map of each task area. 4) Obtain a binary map of the cultivated land prediction results through skeleton line extraction and broken line connection, and finally obtain the cultivated land extraction vector result through raster vectorization and vector straightening, smoothing and other post-processing.

[0081] For edge model extraction:

[0082] The edge model algorithm is divided into two different extraction methods according to different plot texture characteristics: one is for plots with regular morphology, using a conventional edge model; the other is for slender terraces, using a fine edge model.

[0083] The conventional edge model algorithm uses the D-LinkNet network model, which has a high extraction rate of regular cultivated land and a lower leakage rate. The model uses an encoder-decoder structure, dilated convolutions, and pre-trained encoders for extraction tasks. It is built using the LinkNet architecture, which is very efficient in terms of computation and storage. Linknet is an efficient semantic segmentation neural network that has the advantages of skip connections, residual blocks, and encoder-decoder architectures, and has dilated convolution layers in its central part. Dilated convolution is a powerful tool that can expand the perception without reducing the resolution of feature maps. Taking into account aspects such as the connectivity and complexity of the plots, the perception range of the features in the central part of the network is increased, and detailed information is retained.

[0084] The thin edge model algorithm is implemented using edge detection methods. The extracted terrace edges are narrow and not easily confused with boundaries, but the missed rate is high. Edge detection is a fundamental problem in computer vision. In recent years, convolutional neural networks (CNNs) have made significant progress in this field. Existing methods use deep CNNs at a specific level, which may not capture complex data structures due to changes in scale and aspect ratio. In this scheme, an accurate edge detector using richer convolutional features (RCF) is used. RCF encapsulates all convolutional features into a more discriminative representation, fully utilizes the rich feature hierarchy, and can be trained through back-propagation. RCF makes full use of the multi-scale and multi-level information of the target to comprehensively perform image-to-image prediction.

[0085] For texture model extraction:

[0086] The texture model algorithm adopts the D-LinkNet network model in semantic segmentation. Semantic segmentation is a typical computer vision problem that involves taking some raw data (e.g., flat images) as input and converting them into masks with highlighted regions of interest. Many people use the term full-pixel semantic segmentation, where each pixel in the image is assigned a category ID according to the object of interest to which it belongs. Early computer vision problems only found elements such as edges (lines and curves) or gradients, but they never fully provided pixel-level image understanding in the way humans perceive. Semantic segmentation solves this problem by grouping together parts of images belonging to the same target, thereby expanding its application areas. Linknet is an effective semantic segmentation neural network that takes advantage of skip connections, residual blocks, and encoder-decoder structures. D-LinkNet uses Linknet with a pre-trained encoder as its backbone, and has additional dilated convolutional layers in the center part.

[0087] For hierarchical extraction:

[0088] (1)Paddy field

[0089] The edge of the paddy field is affected by the shadows of vegetation, ridges, duckweed, etc., and the outline is not clear. The interior is a typical water feature with uniform water quality, so the texture model is preferred for paddy field extraction. Using the texture model alone will result in incomplete boundaries, so it is necessary to perform a secondary conventional edge model extraction on the paddy field cultivation area. The fusion order is to update and optimize the conventional edge model extraction results with the texture model extraction results, and perform manual verification to finally obtain a map of the paddy field plot.

[0090] (2) Regular planting plots

[0091] Regularly planted plots are mostly distributed in flat land with flat terrain and clear edge contours. The texture of the plot is significantly affected by the type of crops planted, and there are also mixed plantings of multiple crops in the same plot, which leads to complex and diverse textures inside the plot. Therefore, the conventional edge model is preferred for extraction. There are still a few cases where the boundary contours of the regularly planted plots are blurred. In view of this situation, it is necessary to further extract the texture model of the plot and fuse the extraction results. The fusion order is to update the optimized texture model extraction results with the conventional edge model extraction results, and then verify them to obtain a picture of the regularly planted plots.

[0092] (3) Fuzzy planting plots

[0093] Fuzzy planting plots are mostly distributed among mountain forests and grasses, with blurred edge contours and irregular shapes. The texture of the plot is also easily confused with the surrounding weeds, so it is necessary to combine the edge extraction model with the texture extraction model to maximize the extraction rate of fuzzy planting plots. The fusion order is based on the plot characteristics of the fuzzy planting plots. Compared with the edge characteristics, the cultivation type of the fuzzy planting plots is relatively simple and the texture characteristics are more significant. The texture model extraction is preferred, and then the conventional edge model extraction results are updated and optimized.

[0094] (4) Terraced fields

[0095] Terraces are located on hillsides, with clear edge contours and a narrower and longer shape than regular cultivated land. Terraces are planted with a variety of crops and unstable texture homogeneity, so the fine edge extraction model is preferred, supplemented by texture extraction results to optimize the extraction results.

[0096] (5) Garden

[0097] The shape of the garden is irregular, and the edge contour is not prominent. Its notable feature is that the fruit trees are obviously granular and the image texture is clear. The use of texture model extraction has great advantages. The texture model is preferred for extraction, and the conventional edge model extraction results are used as a supplement.

[0098] This embodiment has the following beneficial effects:

[0099] (1) This patent first uses the partitioning idea to partition the data, which can simplify the complex surface and concentrate the elements of similar plot characteristics into several relatively independent sub-areas, reducing the difficulty of extracting objects. At the same time, the data volume of general high-resolution images is large. When the extraction area is large, direct extraction is often time-consuming and laborious, and even impossible to extract, and the hardware requirements are also relatively high. After partitioning, the image with a large amount of data is divided into several small areas, which can effectively disperse the data volume of a single image and reduce the possibility of problems caused by large data volume during the extraction process. At the same time, multi-machine multi-task parallel processing can also be performed to significantly shorten the extraction time.

[0100] (2) By using the layering concept, the surface is divided into types with their own characteristics, and targeted extraction methods are selected according to the specific visual characteristics of each type. Compared with other direct extraction methods without partitioning and stratification, the effect of wrong extraction and missed extraction is significantly improved.

[0101] (3) Different layer fusion schemes are constructed by taking advantage of the differences in terrain and plot characteristics of each farming area. Based on the five major land types of paddy fields, regular planting plots, fuzzy planting plots, terraces, and gardens, different model extraction algorithms are used to propose different layer fusion schemes, and the extraction results with higher accuracy are selected to ensure the quality of the "fine plot" product.

[0102] As another specific implementation process, the process of implementing each step is as follows.

[0103] Image partitioning: Based on the DEM raster data, calculate the terrain data such as slope, aspect, elevation, valley line, ridge line, etc. of the study area, combine the township-level administrative vector data, use the high-resolution remote sensing image as the base map to divide the project construction area, and output the partition surface feature shp file. The partitioning steps are as follows:

[0104] 1) Obtain high-resolution remote sensing images of the target area, such as Google, GF2, QuickBird, etc.;

[0105] 2) Perform image preprocessing such as orthorectification, radiation calibration, atmospheric correction, and cloud removal on the original data;

[0106] 3) Collect administrative data of the target area and adjust administrative division boundaries;

[0107] 4) Collect DEM data of the target area, calculate the surface information, combine the image surface features, and draw the boundaries of the planting area;

[0108] 5) Integrate administrative divisions and planting area boundaries to form a complete zoning control network;

[0109] 6) Analyze the texture characteristics of cultivated land in each planting area and use this as a basis for zoning cultivation;

[0110] The partition control network is used to crop the image to form several sub-task areas with small data volume.

[0111] “Fine plot” map extraction: sample preparation, model training and image prediction, raster vectorization, vector post-processing and vector fusion are carried out in sequence.

[0112] Among them, the sample production part mainly includes five types: paddy fields, regular planting plots, fuzzy planting plots, terraces, and gardens.

[0113] Sample selection criteria:

[0114] 1) Number of samples: more than 100.

[0115] 2) Sample size: 1000*1000 meters.

[0116] 3) Sample selection: Select sample points based on the image features in each sub-area. The samples should be evenly distributed and of moderate quantity.

[0117] Sample preparation steps:

[0118] 1) According to different types of sample points, crop the image to form a fixed size and the corresponding vector frame.

[0119] 2) Perform feature drawing on the cropped sample file. All target objects need to be outlined and matched with the image base map.

[0120] Assign attribute information to the drawn vectors. Attribute annotation distinguishes the types of plots, and converts the samples with attributes into binary images for training.

[0121] The model training and image prediction parts mainly include:

[0122] 1) According to the type of sample, the sample is input into the corresponding convolutional neural network for training;

[0123] 2) Input the images and extraction models of each sub-region into the corresponding neural network, predict the images, and obtain the rule prediction intensity map.

[0124] The raster vectorization and vector post-processing parts mainly include:

[0125] The predicted intensity map is vectorized. The edge intensity map uses relevant tools to extract skeleton lines, connects broken lines and then vectorizes them, and then simplifies and straightens the vector lines and converts them into vector surfaces to obtain the spots; the surface predicted intensity map is converted into a vector surface result after expansion, hole filling and binarization.

[0126] The vector fusion part mainly includes:

[0127] 1) Paddy field farming area: Texture model extraction results are updated and optimized with conventional edge model extraction results;

[0128] 2) Regular planting plots and cultivated areas: the conventional edge model extraction results are updated to optimize the texture model extraction results;

[0129] 3) Fuzzy planting plots and cultivated areas: texture model extraction, and then update and optimize the conventional edge model extraction results;

[0130] 4) Terraced farming areas: extract fine edge models and update and optimize texture extraction results;

[0131] 5) Garden-type cultivated areas: texture model extraction, update and optimize the conventional edge model extraction results.

[0132] 6) Merge the result patches of each cultivated area to obtain the extraction result vectors of each “fine plot” type of the complete area;

[0133] 7) Convert the updated layer from single component to multiple components to eliminate the small fragments and narrow patches, and obtain the final "fine plot" map.

[0134] In addition, the present invention also provides a planting plot spatial structure reconstruction system, comprising:

[0135] Remote sensing data acquisition module, used to obtain remote sensing images of the target planting area;

[0136] A partition structure module is used to partition and deconstruct the remote sensing image to obtain multiple farming areas and corresponding area types; the area types include paddy fields, regular planting plots, fuzzy planting plots, terraces and gardens;

[0137] An edge prediction module is used to select a corresponding image extraction model according to the type of the region, and use the image extraction model to predict the image of each of the farming areas to obtain a prediction intensity map; the prediction intensity map includes an edge prediction intensity map and a surface prediction intensity map; the image extraction model includes a conventional edge model, a fine edge model and a texture model; the conventional edge model and the texture model are constructed based on the D-LinkNet network model; the fine edge model is constructed based on the RCF edge detector;

[0138] A vector processing module, used for performing raster vectorization processing on the predicted intensity map to obtain a spot vector surface;

[0139] The vector fusion module is used to perform vector fusion processing on each of the map spot vector surfaces to obtain the plot map spot data.

[0140] The present invention also provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the above-mentioned method for reconstructing the spatial structure of a planting plot.

[0141] The present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the method for reconstructing the spatial structure of a planting plot as described above.

[0142] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0143] This article uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only used to help understand the core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for reconstructing the spatial structure of a planting plot, It is characterized in that include: Obtain remote sensing images of the target planting area; Deconstructing the remote sensing images by partitioning to obtain a plurality of farming areas and corresponding area types; the area types include paddy fields, regular planting plots, fuzzy planting plots, terraces and gardens; Selecting a corresponding image extraction model according to the type of region, and using the image extraction model to predict the image of each farming area to obtain a prediction intensity map; the prediction intensity map includes an edge prediction intensity map and a surface prediction intensity map; the image extraction model includes a conventional edge model, a fine edge model and a texture model; the conventional edge model and the texture model are constructed based on the D-LinkNet network model; the fine edge model is constructed based on the RCF edge detector; Performing raster vectorization processing on the predicted intensity map to obtain a spot vector surface; Vector fusion processing is performed on each of the above-mentioned map spot vector surfaces to obtain land plot map spot data.

2. The method for reconstructing the spatial structure of a planting plot according to claim 1, It is characterized in that The partitioning and deconstructing according to the remote sensing image to obtain multiple farming areas and corresponding area types specifically includes: Collect administrative data and DEM raster images of the target planting area; Using the remote sensing image as the base map, and based on the administrative data, the "administrative district-planting district-cultivation district-plot" division is carried out to determine the boundaries of the administrative districts; Calculate regional terrain data based on the DEM raster image and draw the boundary of the planting area; Merging the administrative division boundary and the planting area boundary to obtain a division control network; Image clipping is performed based on the regional terrain data and the zoning control network to obtain multiple farming areas and corresponding area types.

3. The method for reconstructing the spatial structure of a planting plot according to claim 2, It is characterized in that Before the remote sensing image is used as the base map and the administrative data is used to divide the "administrative area-planting area-cultivation area-plot" into different areas and the administrative area boundaries are determined, the method further includes: The remote sensing image is subjected to image preprocessing; the image preprocessing includes orthorectification, radiation calibration, atmospheric correction, cloud removal, geometric correction, image enhancement and image mosaicking.

4. The method for reconstructing the spatial structure of a planting plot according to claim 1, It is characterized in that The selecting of a corresponding image extraction model according to the region type and using the image extraction model to predict the image of each farming area to obtain an edge prediction intensity map specifically includes: When the area type is a regular planting plot, the conventional edge model and the texture model are used in turn to predict the edge of the area to obtain an edge prediction intensity map; When the area type is terraced fields, the edge of the area is predicted using a fine edge model and a texture model in turn to obtain an edge prediction intensity map; When the area type is fuzzy planting plots, gardens and paddy fields, the texture model and the conventional edge model are used in turn to perform predictions to obtain an edge prediction intensity map.

5. The method for reconstructing the spatial structure of a planting plot according to claim 1, It is characterized in that The performing grid vectorization processing on the predicted intensity map to obtain a spot vector surface specifically includes: Extracting skeleton lines of the edge prediction strength map, connecting broken lines and then vectorizing them to obtain vector lines; Simplifying and straightening the vector line and then converting it into a first vector surface; After dilation, hole filling and binarization processing are performed on the surface prediction strength map, a second vector surface is obtained; A pattern vector plane is determined according to the first vector plane and the second vector plane.

6. A spatial structure reconstruction system for planting plots, It is characterized in that include: Remote sensing data acquisition module, used to obtain remote sensing images of the target planting area; A partition structure module is used to partition and deconstruct the remote sensing image to obtain multiple farming areas and corresponding area types; the area types include paddy fields, regular planting plots, fuzzy planting plots, terraces and gardens; An edge prediction module is used to select a corresponding image extraction model according to the type of the region, and use the image extraction model to predict the image of each of the farming areas to obtain a prediction intensity map; the prediction intensity map includes an edge prediction intensity map and a surface prediction intensity map; the image extraction model includes a conventional edge model, a fine edge model and a texture model; the conventional edge model and the texture model are constructed based on the D-LinkNet network model; the fine edge model is constructed based on the RCF edge detector; A vector processing module, used for performing raster vectorization processing on the predicted intensity map to obtain a spot vector surface; The vector fusion module is used to perform vector fusion processing on each of the map spot vector surfaces to obtain the plot map spot data.

7. An electronic device, It is characterized in that It comprises a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the method for reconstructing the spatial structure of a planting plot according to any one of claims 1-5.

8. A computer-readable storage medium, It is characterized in that It stores a computer program, which, when executed by a processor, implements the method for reconstructing the spatial structure of a planting plot as described in any one of claims 1-5.

Images