Multi-source data fusion planting plot space structure reconstruction method and system

Abstract

The invention discloses a planting plot space structure reconstruction method and system based on multi-source data fusion, and relates to the technical field of remote sensing application, and the method comprises the steps: obtaining a high-resolution optical image, a synthetic aperture radar image and digital elevation model data of a target region, carrying out the radiometric calibration, atmospheric correction, terrain correction and registration processing, and carrying out the reconstruction of the spatial structure of the planting plot; generating a multi-source data cube; spectral features, texture features, scattering features and topographic features are extracted from the multi-source data cube, channel dimension stacking is carried out, and a high-dimensional multi-channel feature image is generated; a back scattering coefficient graph of the analog synthetic aperture radar is calculated by using a surface microwave scattering physical model and a high-dimensional multi-channel feature image, and a scattering residual image is obtained by calculating a residual with a real synthetic aperture radar image. According to the method, feature aggregation and semantic similarity calculation are carried out through a graph convolutional neural network, graph spots are optimized and merged according to a node incidence matrix, and a high-precision plot space structure reconstruction vector graph is output after topology check and geometric correction.

Description

Technical Field

[0001] This invention relates to the field of remote sensing application technology, and in particular to a method and system for reconstructing the spatial structure of planting plots through multi-source data fusion. Background Technology

[0002] In recent years, with the rapid development of remote sensing technology, the method of extracting planting plots based on high-resolution optical images has become an important research direction in the field of agricultural remote sensing. Traditional methods mainly rely on spectral features and texture analysis, and identify plot boundaries through object-oriented classification, edge detection algorithms or machine learning models. By fusing visible vegetation index and texture features, it can achieve high-precision plot division in plain areas. With the introduction of deep learning technology, semantic segmentation models such as U-Net and DeepLab have been widely used in remote sensing image plot extraction tasks. These models learn spatial features in images through encoder-decoder structures, which significantly improves the boundary recognition ability of regular planting areas.

[0003] However, the above methods still face significant challenges in complex planting areas. Since optical images are easily affected by clouds, shadows and changes in vegetation cover, it is difficult to reliably identify planting plots with blurred boundaries by relying solely on spectral-texture features. Although existing multi-source fusion methods integrate optical and radar data, they usually adopt simple data stacking or weighted fusion strategies, which fail to fundamentally solve the problem of physical consistency between different modal data. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for reconstructing the spatial structure of planting plots by fusing multi-source data, which solves the problems of feature inconsistency and plot fragmentation caused by differences in the physical properties of data in complex environments.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for reconstructing the spatial structure of planting plots by fusing multi-source data, which includes acquiring high-resolution optical images, synthetic aperture radar images and digital elevation model data of the target area, performing radiometric calibration, atmospheric correction, terrain correction and registration processing, and generating a multi-source data cube. Spectral features, texture features, scattering features and terrain features are extracted from a multi-source data cube, and channel dimensions are stacked to generate a high-dimensional multi-channel feature image. The backscattering coefficient map of simulated synthetic aperture radar is calculated using the physical model of surface microwave scattering and high-dimensional multi-channel feature images. The residual is calculated with the real synthetic aperture radar image to obtain the scattering residual image. The image is then stacked with the high-dimensional multi-channel feature image and input into a deep learning network for feature fusion to generate a fused feature map with enhanced land parcel boundaries. The enhanced fusion feature map of the land parcel boundaries is input into a pre-trained convolutional neural network boundary prediction model to generate a land parcel boundary probability map. Thresholding and morphological refinement are performed on the probability map of the land parcel boundary to extract the skeleton lines. After disconnection and polygonization, an initial vector pattern set is generated. The initial set of vector map features is converted into graph structure data, which is then input into a graph convolutional neural network for feature aggregation and semantic similarity calculation. Based on the nodes, the association matrix is ​​optimized by merging map features to generate an optimized plot vector surface dataset. The optimized plot vector surface dataset is then subjected to topological checking and geometric correction. After eliminating overlapping gaps, it is assigned an identifier and the final plot spatial structure reconstruction vector map is output.

[0007] As a preferred embodiment of the multi-source data fusion method for reconstructing the spatial structure of planting plots according to the present invention, the method includes the following steps: acquiring high-resolution optical images, synthetic aperture radar images, and digital elevation model data of the target area; performing radiometric calibration, atmospheric correction, terrain correction, and registration processing to generate a multi-source data cube. Acquire high-resolution optical images, synthetic aperture radar images, and digital elevation model data of the target area, and perform radiometric calibration on the high-resolution optical images, synthetic aperture radar images, and digital elevation model data; Atmospheric correction is performed on the radiometrically calibrated high-resolution optical images, synthetic aperture radar images, and digital elevation model data. Topographic correction is performed on the atmospherically corrected high-resolution optical images, synthetic aperture radar images, and digital elevation model data. The high-resolution optical imagery, synthetic aperture radar imagery, and digital elevation model data after terrain correction are registered, and a spatially matched multi-source data cube is generated through the registration process.

[0008] As a preferred embodiment of the multi-source data fusion method for reconstructing the spatial structure of planting plots according to the present invention, the method includes the following steps: extracting spectral features, texture features, scattering features, and topographic features from a multi-source data cube, stacking the channel dimensions, and generating a high-dimensional multi-channel feature image. Spectral features, texture features, scattering features, and terrain features are extracted from a multi-source data cube. The spectral features, texture features, scattering features, and terrain features are then stacked along the channel dimension to generate a high-dimensional multi-channel feature image.

[0009] As a preferred embodiment of the multi-source data fusion method for reconstructing the spatial structure of planting plots according to the present invention, the method includes: calculating a simulated synthetic aperture radar backscattering coefficient map using a surface microwave scattering physical model and high-dimensional multi-channel feature images; calculating the residual between the simulated synthetic aperture radar image and the real synthetic aperture radar image to obtain a scattering residual image; stacking the residual image with the high-dimensional multi-channel feature images and inputting it into a deep learning network for feature fusion to generate a fused feature map with enhanced plot boundaries, comprising the following steps: The backscattering coefficient map of simulated synthetic aperture radar is calculated using the physical model of surface microwave scattering and high-dimensional multi-channel feature images. The scattering residual image is obtained by calculating the residual between the simulated synthetic aperture radar backscattering coefficient map and the real synthetic aperture radar image. The scattering residual image and the high-dimensional multi-channel feature image are stacked according to channel dimensions; the stacked features are input into a deep learning network for feature fusion, and a fused feature map with enhanced land parcel boundaries is generated through feature fusion.

[0010] As a preferred embodiment of the multi-source data fusion method for reconstructing the spatial structure of planting plots according to the present invention, the method includes the following steps: inputting the enhanced fusion feature map of plot boundaries into a pre-trained convolutional neural network boundary prediction model to generate a plot boundary probability map. The enhanced fusion feature map of the land parcel boundaries is input into a pre-trained convolutional neural network boundary prediction model. The pre-trained convolutional neural network boundary prediction model performs forward propagation on the enhanced fusion feature map of the land parcel boundaries to generate a land parcel boundary probability map.

[0011] As a preferred embodiment of the multi-source data fusion method for reconstructing the spatial structure of planting plots according to the present invention, the method includes the following steps: thresholding and morphological refinement of the plot boundary probability map to extract skeleton lines, followed by line disconnection and polygonization to generate an initial vector pattern set. Thresholding is performed on the probability map of the land parcel boundary to obtain a binarized boundary image. Morphological thinning is performed on the binarized boundary image to extract the skeleton line. The skeleton line is then broken and connected to obtain a continuous and complete boundary line network. The continuous and complete boundary line network is polygonized; the initial vector pattern set is generated through polygonization.

[0012] As a preferred embodiment of the multi-source data fusion method for reconstructing the spatial structure of planting plots according to the present invention, the method includes the following steps: converting the initial vector patch set into graph structure data, inputting it into a graph convolutional neural network for feature aggregation and semantic similarity calculation, optimizing the patch merging based on nodes in the association matrix, generating an optimized plot vector surface dataset, performing topological checks and geometric corrections on the optimized plot vector surface dataset, eliminating overlapping gaps and assigning identifiers, and outputting the final plot spatial structure reconstruction vector map. The initial set of vector patches is converted into graph structure data, and edge connections are established between adjacent patches. The graph structure data is then input into a graph convolutional neural network for node feature aggregation and semantic similarity calculation. A node pair association matrix is ​​generated through a graph convolutional neural network. Adjacent parcels are merged and optimized based on the node pair association matrix. An optimized parcel vector surface dataset is generated through the merging and optimization process. A topology check is performed on the optimized parcel vector surface dataset after the topology check. Geometric errors are corrected in the optimized parcel vector surface dataset after the topology check. After correcting geometric errors, the optimized plot vector surface dataset is de-overlapped and gapped. A unique identifier is assigned to the optimized plot vector surface dataset after eliminating overlaps and gaps. The final high-precision plot spatial structure reconstruction vector map is then output by assigning a unique identifier.

[0013] Secondly, the present invention provides a multi-source data fusion system for reconstructing the spatial structure of planting plots, including a data acquisition module, which acquires high-resolution optical images, synthetic aperture radar images and digital elevation model data of the target area, performs radiometric calibration, atmospheric correction, terrain correction and registration processing, and generates a multi-source data cube. The extraction module extracts spectral features, texture features, scattering features and terrain features from the multi-source data cube, stacks them in channel dimensions, and generates a high-dimensional multi-channel feature image. The fusion module uses a physical model of surface microwave scattering and high-dimensional multi-channel feature images to calculate a simulated synthetic aperture radar backscattering coefficient map. It calculates the residual with the real synthetic aperture radar image to obtain a scattering residual image. After stacking with the high-dimensional multi-channel feature image, it is input into a deep learning network for feature fusion to generate a fused feature map with enhanced land parcel boundaries. The prediction module inputs the enhanced fusion feature map of the land parcel boundaries into a pre-trained convolutional neural network boundary prediction model to generate a land parcel boundary probability map. The threshold processing module performs threshold processing and morphological refinement on the probability map of land parcel boundaries to extract skeleton lines. After disconnection and polygonization, it generates an initial set of vector pattern pieces. The optimization module converts the initial vector patch set into graph structure data, inputs it into a graph convolutional neural network for feature aggregation and semantic similarity calculation, optimizes the patch merging based on the node association matrix, generates an optimized plot vector surface dataset, performs topological checks and geometric corrections on the optimized plot vector surface dataset, eliminates overlapping gaps and assigns identifiers, and outputs the final plot spatial structure reconstruction vector map.

[0014] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the planting plot spatial structure reconstruction method of multi-source data fusion as described in the first aspect of the present invention.

[0015] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the method for reconstructing the spatial structure of planting plots by multi-source data fusion as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: By acquiring high-resolution optical images, synthetic aperture radar images, and digital elevation model data, a multi-source data cube is generated through radiometric calibration, atmospheric correction, terrain correction, and registration. Spectral, texture, scattering, and terrain features are extracted from the cube and stacked to generate a high-dimensional multi-channel feature image. A simulated backscattering coefficient map is calculated using a surface microwave scattering physical model, and the scattering residual image is obtained by calculating the residual with the real image. This residual image is then stacked with the high-dimensional features and fused through a deep learning network to generate a fused feature map that enhances the boundaries of the land parcels. This fused feature map is then input into a pre-trained convolutional neural network to generate a probability map of the land parcel boundaries. After thresholding, morphological thinning, and disconnection, the map is polygonized to generate an initial set of vector map features. Finally, the set of map features is converted into graph structure data, and feature aggregation and semantic similarity calculation are performed through a graph convolutional neural network. The map features are optimized and merged based on the node association matrix. After topological checking and geometric correction, a high-precision vector map of the reconstructed spatial structure of the land parcels is output. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart for a method of reconstructing the spatial structure of planting plots using multi-source data fusion.

[0019] Figure 2A schematic diagram of a planting plot spatial structure reconstruction system based on multi-source data fusion. Detailed Implementation

[0020] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention; however, the present invention may also be implemented in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0021] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0022] Reference Figures 1-2 As one embodiment of the present invention, this embodiment provides a method for reconstructing the spatial structure of planting plots through multi-source data fusion, comprising the following steps: S1. Acquire high-resolution optical images, synthetic aperture radar images, and digital elevation model data of the target area, perform radiometric calibration, atmospheric correction, terrain correction, and registration processing, and generate a multi-source data cube.

[0023] S1.1 Acquire high-resolution optical images, synthetic aperture radar images, and digital elevation model data of the target area, and perform radiometric calibration on the high-resolution optical images, synthetic aperture radar images, and digital elevation model data.

[0024] Furthermore, radiometric calibration of high-resolution optical imagery is accomplished by converting the raw digital quantization values ​​of the imagery into apparent reflectance or surface reflectance. Radiometric calibration of synthetic aperture radar imagery converts backscattering intensity values ​​into normalized backscattering coefficients by applying calibration parameters provided by the sensor. Radiometric calibration of digital elevation model data involves checking and ensuring the consistency and validity of the units of measurement for elevation values.

[0025] S1.2 Perform atmospheric correction on the radiometrically calibrated high-resolution optical images, synthetic aperture radar images, and digital elevation model data, and perform terrain correction on the atmospherically corrected high-resolution optical images, synthetic aperture radar images, and digital elevation model data.

[0026] Furthermore, atmospheric correction of the radiometrically calibrated high-resolution optical images employs methods based on radiative transfer models, such as the 6S model or the MODTRAN model, to eliminate the effects of scattering and absorption by atmospheric molecules and aerosols. Atmospheric correction of the radiometrically calibrated synthetic aperture radar (SAR) images primarily targets ionospheric and tropospheric delay effects, using correction methods based on external meteorological data or models. After atmospheric correction of both the high-resolution optical images and SAR images, topographic correction is performed on the high-resolution optical images using digital elevation model data to eliminate pixel brightness variations caused by terrain undulations, and topographic correction is performed on the SAR images to compensate for geometric deformation and radiometric distortion.

[0027] S1.3. The high-resolution optical imagery, synthetic aperture radar imagery, and digital elevation model data after terrain correction are registered to generate a spatially matched multi-source data cube.

[0028] Furthermore, spatial registration is performed on the terrain-corrected high-resolution optical images, synthetic aperture radar images, and digital elevation model data using feature point matching or region matching algorithms. This ensures that the pixels from all data sources are precisely aligned in a unified geographic coordinate system, ultimately generating a spatially matched multi-source data cube.

[0029] S2. Extract spectral features, texture features, scattering features and terrain features from the multi-source data cube, stack the channel dimensions, and generate a high-dimensional multi-channel feature image.

[0030] S2.1 Extract spectral features from the multi-source data cube, extract texture features from the multi-source data cube, extract scattering features from the multi-source data cube, extract terrain features from the multi-source data cube, stack the spectral features, texture features, scattering features and terrain features by channel dimension, and generate a high-dimensional multi-channel feature image by channel dimension stacking.

[0031] Furthermore, spectral features are extracted from the multi-source data cube by calculating vegetation indices such as the Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EDI) for each band of the high-resolution optical image; texture features are extracted from the multi-source data cube by using the gray-level co-occurrence matrix algorithm to obtain statistical quantities such as contrast, entropy, and second moment of the high-resolution optical image; scattering features are extracted from the multi-source data cube by deriving parameters such as mean, variance, and isopolarization ratio from the backscattering coefficients of the synthetic aperture radar image; topographic features are extracted from the multi-source data cube by deriving information such as slope, aspect, and topographic relief from digital elevation model data; channel-dimensional stacking of spectral features, texture features, scattering features, and topographic features involves stitching all the obtained spectral feature maps, texture feature maps, scattering feature maps, and topographic feature maps into a multidimensional data volume according to channel order; the high-dimensional multi-channel feature image generated by channel-dimensional stacking contains all the extracted feature information as input for subsequent processing.

[0032] S3. Calculate the simulated synthetic aperture radar backscattering coefficient map using the surface microwave scattering physical model and high-dimensional multi-channel feature images. Calculate the residual with the real synthetic aperture radar image to obtain the scattering residual image. Stack the residual with the high-dimensional multi-channel feature image and input it into a deep learning network for feature fusion to generate a fused feature map with enhanced land parcel boundaries.

[0033] S3.1 Calculate the simulated synthetic aperture radar backscattering coefficient map using the surface microwave scattering physical model and high-dimensional multi-channel feature image. Calculate the residual between the simulated synthetic aperture radar backscattering coefficient map and the real synthetic aperture radar image to obtain the scattering residual image.

[0034] The expression for the scattering residual image is: ; in, For the scattering residual image, To provide a realistic synthetic aperture radar image, is the numerical stability constant.

[0035] Furthermore, the specific implementation method for simulating the synthetic aperture radar backscattering coefficient map using the surface microwave scattering physical model and high-dimensional multi-channel feature images is as follows: spectral features and topographic features from the high-dimensional multi-channel feature images are input into the surface microwave scattering physical model. A simulated synthetic aperture radar backscattering coefficient map with the same spatial resolution as the actual synthetic aperture radar image is generated through forward simulation calculation of the physical model. When calculating the residual between the simulated synthetic aperture radar backscattering coefficient map and the actual synthetic aperture radar image, a pixel-by-pixel absolute difference calculation is used to obtain the scattering residual image. Areas with larger values ​​in the scattering residual image represent surface areas where there is a significant difference between the physical model prediction and the actual observation.

[0036] S3.2 Stack the scattering residual image with the high-dimensional multi-channel feature image according to the channel dimension; input the stacked features into a deep learning network for feature fusion, and generate a fused feature map with enhanced land parcel boundaries through feature fusion.

[0037] Furthermore, the specific implementation of stacking the scattering residual image with the high-dimensional multi-channel feature image in terms of channel dimensions is as follows: the scattering residual image is used as a new feature channel and stitched together with the original feature channels of the high-dimensional multi-channel feature image; when the stacked features are input into a deep learning network for feature fusion, a convolutional neural network with an encoder-decoder structure is used to learn the nonlinear relationship between features through multi-layer convolution and deconvolution operations, and finally generate a fused feature map with enhanced land parcel boundaries.

[0038] S4. Input the enhanced fusion feature map of the land parcel boundary into the pre-trained convolutional neural network boundary prediction model to generate a land parcel boundary probability map.

[0039] S4.1 Input the enhanced fusion feature map of the land parcel boundary into the pre-trained convolutional neural network boundary prediction model. The pre-trained convolutional neural network boundary prediction model performs forward propagation on the enhanced fusion feature map of the land parcel boundary to generate a land parcel boundary probability map.

[0040] Furthermore, the specific implementation of inputting the enhanced fusion feature map of land parcel boundaries into the pre-trained convolutional neural network boundary prediction model is as follows: The pre-trained convolutional neural network boundary prediction model adopts an encoder-decoder architecture. The encoder part gradually extracts multi-level feature representations from the enhanced fusion feature map of land parcel boundaries through multi-layer convolution and pooling operations. The decoder part gradually restores the spatial resolution of the feature map through deconvolution and upsampling operations and outputs a probability map with the same size as the input. When the pre-trained convolutional neural network boundary prediction model performs forward propagation on the enhanced fusion feature map of land parcel boundaries, it sequentially performs linear transformations of convolutional layers and non-linear transformations of activation functions. Finally, it generates the probability value of each pixel belonging to the land parcel boundary through the sigmoid activation function, forming a land parcel boundary probability map. The value of each pixel in the land parcel boundary probability map ranges from zero to one. The closer the value is to one, the higher the probability that the position is the land parcel boundary. The closer the value is to zero, the higher the probability that the position is inside the land parcel.

[0041] S5. Thresholding and morphological refinement are performed on the probability map of the land parcel boundaries to extract the skeleton lines. After disconnection and polygonization, an initial vector pattern set is generated.

[0042] S5.1. Thresholding is performed on the probability map of the land parcel boundary to obtain a binarized boundary image. Morphological thinning is performed on the binarized boundary image to extract the skeleton lines. The skeleton lines are then broken and connected to obtain a continuous and complete boundary line network.

[0043] Furthermore, regions with pixel values ​​greater than a set threshold in the land parcel boundary probability map are classified as boundary pixels and assigned a value of one, while regions with pixel values ​​less than or equal to the set threshold are classified as non-boundary pixels and assigned a value of zero, thereby generating a binary boundary image containing only black and white pixel values. When performing morphological thinning operations on the binary boundary image to extract the skeleton line, an iterative erosion algorithm is used to gradually remove boundary pixels until only a single-pixel-width centerline structure is retained. This centerline structure is the skeleton line reflecting the direction of the land parcel boundary. When performing broken line connection processing on the skeleton line to obtain a continuous and complete boundary line network, a gap connection method based on morphological dilation is used to expand local pixels at the broken points of the skeleton line to connect the broken endpoints, ultimately forming a spatially continuous and unbroken boundary line network.

[0044] S5.2. Perform polygonization on the continuous and complete boundary line network; generate an initial vector pattern set through polygonization.

[0045] Furthermore, a boundary tracing algorithm is used to identify the coordinate sequences of all closed loops from a continuous and complete boundary line network. For boundary segments that are not completely closed, an endpoint connection strategy is used to force them to form closed polygons. The coordinate sequence of each closed loop is converted into a vector polygon feature, and a unique identifier is assigned to each polygon. Through polygonization processing, an initial vector patch set consisting of multiple independent vector polygon features is generated, where each vector polygon feature represents a preliminarily identified planting plot unit.

[0046] S6. Convert the initial vector map patch set into graph structure data, input it into the graph convolutional neural network for feature aggregation and semantic similarity calculation, optimize the patch merging based on the nodes in the association matrix, generate an optimized plot vector surface dataset, perform topological checking and geometric correction on the optimized plot vector surface dataset, eliminate overlapping gaps and assign identifiers, and output the final plot spatial structure reconstruction vector map.

[0047] S6.1. Convert the initial vector patch set into graph structure data, establish edge connections between adjacent patches, and input the graph structure data into a graph convolutional neural network for node feature aggregation and semantic similarity calculation; The similarity score expression is: ; in, The semantic similarity score is given between the first and second image patches. For the first patch and its evolution The final node embedding vector obtained after the layered graph convolutional neural network. The process of the second map patch compared to the first map patch The final node embedding vector obtained after the layered graph convolutional neural network.

[0048] Furthermore, each initial vector patch is treated as a node in the graph structure data. The node feature vector consists of the multi-source feature statistics and morphological features corresponding to the patch. If two initial vector patches are spatially adjacent, an edge connection is established between the corresponding nodes to form graph structure data representing the spatial adjacency relationship of the patches. When the graph structure data is input into the graph convolutional neural network for node feature aggregation, the node features are iteratively updated through multi-layer graph convolution operations, enabling each node to aggregate the feature information of its neighboring nodes. Semantic similarity is calculated using the cosine similarity method, which calculates the cosine value of the angle between the final node embedding vectors obtained after processing by the graph convolutional neural network, to obtain a similarity score reflecting the degree of semantic similarity between patches.

[0049] S6.2 Generate a node pair association matrix using a graph convolutional neural network, perform merging and optimization processing on adjacent plots based on the node pair association matrix, generate an optimized plot vector surface dataset through merging and optimization processing, perform topology checking on the optimized plot vector surface dataset, and perform geometric error correction on the optimized plot vector surface dataset after topology checking.

[0050] Furthermore, a node pair association matrix is ​​constructed based on the semantic similarity scores between all adjacent patch node pairs, where each element represents the merging probability of the corresponding patch pair. When merging adjacent patches based on the node pair association matrix, a probability threshold judgment method is used to perform vector surface merging operations on patch pairs whose merging probability exceeds a set threshold. An optimized land parcel vector surface dataset is generated through merging optimization, which reduces the number of patches and improves land parcel integrity. A topology check is performed on the optimized land parcel vector surface dataset to detect and record the types of topology errors. Geometric error correction is performed on the optimized land parcel vector surface dataset after topology check, including removing duplicate vertices and correcting geometric problems such as loop self-intersections.

[0051] S6.3. After correcting geometric errors, the optimized plot vector surface dataset is de-overlapped and gapped. A unique identifier is assigned to the optimized plot vector surface dataset after eliminating overlaps and gaps. The final high-precision planting plot spatial structure reconstruction vector map is output by assigning a unique identifier.

[0052] Furthermore, buffer analysis and overlay analysis methods are used to identify and eliminate overlapping areas between vector surfaces, while gap detection algorithms are used to fill the gaps between vector surfaces. The optimized plot vector surface dataset after eliminating overlaps and gaps is assigned a unique identifier, and each independent plot polygon is assigned a unique number. By assigning unique identifiers, the final high-precision planting plot spatial structure reconstruction vector map is output, which contains a complete set of plot polygons without topological errors and with unique identifiers.

[0053] This embodiment also provides a multi-source data fusion system for reconstructing the spatial structure of planting plots, including: a data acquisition module, which acquires high-resolution optical images, synthetic aperture radar images and digital elevation model data of the target area, performs radiometric calibration, atmospheric correction, terrain correction and registration processing, and generates a multi-source data cube; The extraction module extracts spectral features, texture features, scattering features and terrain features from the multi-source data cube, stacks them in channel dimensions, and generates a high-dimensional multi-channel feature image. The fusion module uses a physical model of surface microwave scattering and high-dimensional multi-channel feature images to calculate a simulated synthetic aperture radar backscattering coefficient map. It calculates the residual with the real synthetic aperture radar image to obtain a scattering residual image. After stacking with the high-dimensional multi-channel feature image, it is input into a deep learning network for feature fusion to generate a fused feature map with enhanced land parcel boundaries. The prediction module inputs the enhanced fusion feature map of the land parcel boundaries into a pre-trained convolutional neural network boundary prediction model to generate a land parcel boundary probability map. The threshold processing module performs threshold processing and morphological refinement on the probability map of land parcel boundaries to extract skeleton lines. After disconnection and polygonization, it generates an initial set of vector pattern pieces. The optimization module converts the initial vector patch set into graph structure data, inputs it into a graph convolutional neural network for feature aggregation and semantic similarity calculation, optimizes the patch merging based on the node association matrix, generates an optimized plot vector surface dataset, performs topological checks and geometric corrections on the optimized plot vector surface dataset, eliminates overlapping gaps and assigns identifiers, and outputs the final plot spatial structure reconstruction vector map.

[0054] This embodiment also provides a computer device applicable to the method for reconstructing the spatial structure of planting plots using multi-source data fusion, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the method for reconstructing the spatial structure of planting plots using multi-source data fusion as proposed in the above embodiment.

[0055] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0056] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the method for reconstructing the spatial structure of planting plots by fusing multi-source data as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0057] In summary, this invention acquires high-resolution optical images, synthetic aperture radar images, and digital elevation model data, and generates a multi-source data cube through radiometric calibration, atmospheric correction, topographic correction, and registration. Spectral, texture, scattering, and topographic features are extracted from this cube and stacked to generate a high-dimensional multi-channel feature image. A simulated backscattering coefficient map is calculated using a surface microwave scattering physical model, and the residual image is calculated from the residual image to obtain a scattering residual image. This residual image is then stacked with the high-dimensional features and fused using a deep learning network to generate a fused feature map that enhances the boundaries of the land parcels. This fused feature map is then input into a pre-trained convolutional neural network to generate a probability map of the land parcel boundaries. After thresholding, morphological thinning, and disconnection, the map is polygonized to generate an initial set of vector map features. Finally, the set of map features is converted into graph-structured data, and feature aggregation and semantic similarity calculation are performed using a graph convolutional neural network. The map features are then optimized and merged based on the node association matrix. After topological checking and geometric correction, a high-precision vector map of the reconstructed spatial structure of the land parcels is output.

[0058] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for reconstructing the spatial structure of planting plots through multi-source data fusion, characterized in that: This includes acquiring high-resolution optical images, synthetic aperture radar images, and digital elevation model data of the target area, performing radiometric calibration, atmospheric correction, terrain correction, and registration processing, and generating a multi-source data cube; Spectral features, texture features, scattering features and terrain features are extracted from a multi-source data cube, and channel dimensions are stacked to generate a high-dimensional multi-channel feature image. The backscattering coefficient map of simulated synthetic aperture radar is calculated using the physical model of surface microwave scattering and high-dimensional multi-channel feature images. The residual is calculated with the real synthetic aperture radar image to obtain the scattering residual image. The image is then stacked with the high-dimensional multi-channel feature image and input into a deep learning network for feature fusion to generate a fused feature map with enhanced land parcel boundaries. The enhanced fusion feature map of the land parcel boundaries is input into a pre-trained convolutional neural network boundary prediction model to generate a land parcel boundary probability map. Thresholding and morphological refinement are performed on the probability map of the land parcel boundary to extract the skeleton lines. After disconnection and polygonization, an initial set of vector pattern pieces is generated. The initial set of vector map features is converted into graph structure data, which is then input into a graph convolutional neural network for feature aggregation and semantic similarity calculation. Based on the nodes, the association matrix is ​​optimized by merging map features to generate an optimized plot vector surface dataset. The optimized plot vector surface dataset is then subjected to topological checking and geometric correction. After eliminating overlapping gaps, it is assigned an identifier and the final plot spatial structure reconstruction vector map is output.

2. The method for reconstructing the spatial structure of planting plots by multi-source data fusion as described in claim 1, characterized in that: Acquire high-resolution optical imagery, synthetic aperture radar imagery, and digital elevation model data of the target area; perform radiometric calibration, atmospheric correction, terrain correction, and registration processing to generate a multi-source data cube, including the following steps: Acquire high-resolution optical images, synthetic aperture radar images, and digital elevation model data of the target area, and perform radiometric calibration on the high-resolution optical images, synthetic aperture radar images, and digital elevation model data; Atmospheric correction is performed on the radiometrically calibrated high-resolution optical images, synthetic aperture radar images, and digital elevation model data. Topographic correction is performed on the atmospherically corrected high-resolution optical images, synthetic aperture radar images, and digital elevation model data. The high-resolution optical imagery, synthetic aperture radar imagery, and digital elevation model data after terrain correction are registered, and a spatially matched multi-source data cube is generated through the registration process.

3. The method for reconstructing the spatial structure of planting plots by multi-source data fusion as described in claim 2, characterized in that: Extracting spectral features, texture features, scattering features, and topographic features from a multi-source data cube, and stacking them according to channel dimensions to generate a high-dimensional multi-channel feature image, includes the following steps: Spectral features, texture features, scattering features, and terrain features are extracted from a multi-source data cube. The spectral features, texture features, scattering features, and terrain features are then stacked along the channel dimension to generate a high-dimensional multi-channel feature image.

4. The method for reconstructing the spatial structure of planting plots by multi-source data fusion as described in claim 3, characterized in that: The simulated synthetic aperture radar backscattering coefficient map is calculated using a physical model of surface microwave scattering and high-dimensional multi-channel feature images. The residual is calculated with the real synthetic aperture radar image to obtain a scattering residual image. This residual image is then stacked with the high-dimensional multi-channel feature image and input into a deep learning network for feature fusion to generate a fused feature map with enhanced land parcel boundaries. The process includes the following steps: The backscattering coefficient map of simulated synthetic aperture radar is calculated using the physical model of surface microwave scattering and high-dimensional multi-channel feature images. The scattering residual image is obtained by calculating the residual between the simulated synthetic aperture radar backscattering coefficient map and the real synthetic aperture radar image. The scattering residual image and the high-dimensional multi-channel feature image are stacked according to channel dimensions; the stacked features are input into a deep learning network for feature fusion, and a fused feature map with enhanced land parcel boundaries is generated through feature fusion.

5. The method for reconstructing the spatial structure of planting plots by multi-source data fusion as described in claim 4, characterized in that: The enhanced fused feature map of land parcel boundaries is input into a pre-trained convolutional neural network boundary prediction model to generate a land parcel boundary probability map, including the following steps: The enhanced fusion feature map of the land parcel boundaries is input into a pre-trained convolutional neural network boundary prediction model. The pre-trained convolutional neural network boundary prediction model performs forward propagation on the enhanced fusion feature map of the land parcel boundaries to generate a land parcel boundary probability map.

6. The method for reconstructing the spatial structure of planting plots by multi-source data fusion as described in claim 5, characterized in that: The probability map of land parcel boundaries is thresholded and morphologically refined to extract skeleton lines. After disconnection and polygonization, an initial set of vector plots is generated, including the following steps: Thresholding is performed on the probability map of the land parcel boundary to obtain a binarized boundary image. Morphological thinning is performed on the binarized boundary image to extract the skeleton line. The skeleton line is then broken and connected to obtain a continuous and complete boundary line network. The continuous and complete boundary line network is polygonized; the initial vector pattern set is generated through polygonization.

7. The method for reconstructing the spatial structure of planting plots by multi-source data fusion as described in claim 6, characterized in that: The initial vector map patch set is converted into graph structure data, input into a graph convolutional neural network for feature aggregation and semantic similarity calculation, and the patch merging optimization is performed on the association matrix based on nodes to generate an optimized plot vector surface dataset. The optimized plot vector surface dataset undergoes topological checking and geometric correction, overlapping gaps are eliminated, and identifiers are assigned. The final plot spatial structure reconstruction vector map is then output, including the following steps: The initial set of vector patches is converted into graph structure data, and edge connections are established between adjacent patches. The graph structure data is then input into a graph convolutional neural network for node feature aggregation and semantic similarity calculation. A node pair association matrix is ​​generated through a graph convolutional neural network. Adjacent parcels are merged and optimized based on the node pair association matrix. An optimized parcel vector surface dataset is generated through the merging and optimization process. A topology check is performed on the optimized parcel vector surface dataset after the topology check. Geometric errors are corrected in the optimized parcel vector surface dataset after the topology check. After correcting geometric errors, the optimized plot vector surface dataset is de-overlapped and gapped. A unique identifier is assigned to the optimized plot vector surface dataset after eliminating overlaps and gaps. The final high-precision plot spatial structure reconstruction vector map is then output by assigning a unique identifier.

8. A system for reconstructing the spatial structure of planting plots based on multi-source data fusion, wherein the method for reconstructing the spatial structure of planting plots based on multi-source data fusion as described in any one of claims 1 to 7 is characterized in that: This includes a data acquisition module that acquires high-resolution optical images, synthetic aperture radar images, and digital elevation model data of the target area, performs radiometric calibration, atmospheric correction, terrain correction, and registration processing, and generates a multi-source data cube. The extraction module extracts spectral features, texture features, scattering features and terrain features from the multi-source data cube, stacks them in channel dimensions, and generates a high-dimensional multi-channel feature image. The fusion module uses a physical model of surface microwave scattering and high-dimensional multi-channel feature images to calculate a simulated synthetic aperture radar backscattering coefficient map. It calculates the residual with the real synthetic aperture radar image to obtain a scattering residual image. After stacking with the high-dimensional multi-channel feature image, it is input into a deep learning network for feature fusion to generate a fused feature map with enhanced land parcel boundaries. The prediction module inputs the enhanced fusion feature map of the land parcel boundaries into a pre-trained convolutional neural network boundary prediction model to generate a land parcel boundary probability map. The threshold processing module performs threshold processing and morphological refinement on the probability map of land parcel boundaries to extract skeleton lines. After disconnection and polygonization, it generates an initial set of vector pattern pieces. The optimization module converts the initial vector patch set into graph structure data, inputs it into a graph convolutional neural network for feature aggregation and semantic similarity calculation, optimizes the patch merging based on the node association matrix, generates an optimized plot vector surface dataset, performs topological checks and geometric corrections on the optimized plot vector surface dataset, eliminates overlapping gaps and assigns identifiers, and outputs the final plot spatial structure reconstruction vector map.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the planting plot spatial structure reconstruction method based on multi-source data fusion as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the planting plot spatial structure reconstruction method based on multi-source data fusion as described in any one of claims 1 to 7.

Images