Large-scale farmland boundary identification method and system based on multi-source remote sensing data fusion

Abstract

The invention relates to the technical field of object recognition, and discloses a large-scale farmland boundary recognition method and system based on multi-source remote sensing data fusion, and the method comprises the steps: carrying out the adaptive spectrum fusion of a multispectral remote sensing image, and obtaining a spectrum fusion remote sensing image; mapping pixel values of pixels in the spectrum fusion remote sensing image to three-dimensional coordinate points on the surface of the three-dimensional terrain structure model; performing multi-mode gating fusion on the mapped multi-source remote sensing image to obtain a boundary probability image; and extracting a coarse-grained farmland boundary from the boundary probability image, and performing constraint optimization on the coarse-grained farmland boundary in combination with gradient information of pixels in the gradient image. According to the method, a farmland three-dimensional terrain structure model is constructed, SAR polarization characteristics and spectral indexes are fused, a boundary probability graph is generated through multi-modal gating, a coarse-grained boundary is constructed, gradient driving constraints are introduced in combination with terrain gradient information, physical consistency optimization is carried out on the coarse-grained boundary, and a continuous and real farmland boundary is recognized.

Description

Technical Field

[0001] The present invention relates to the technical field of object recognition, in particular to the technical field of farmland boundary recognition, and specifically to a large-scale farmland boundary recognition method and system by fusing multi-source remote sensing data. Background Art

[0002] In agricultural production, boundary identification is not only crucial for monitoring farmland area, analyzing crop growth, and assessing disasters, but also plays a fundamental role in the implementation of precision agriculture. Remote sensing data, due to its high efficiency and large-scale monitoring capabilities, has become a crucial data source for farmland boundary identification research.

[0003] In the field of agricultural remote sensing, high-precision identification of farmland boundaries is of great significance for cultivated land management, land parcel segmentation and agricultural informatization. In existing research, CN114648704A provides a high-precision farmland boundary extraction method and system, including: obtaining remote sensing images of multiple bands of the area to be extracted; inputting the remote sensing images of each band into a cyclic residual convolutional neural network to obtain multiple rough farmland boundary maps; after fusing all the rough farmland boundary maps, calculating the boundary strength in the fused image to obtain a primary farmland boundary map; using the watershed transform algorithm to close the gaps in the boundaries in the primary farmland boundary map to obtain the final farmland boundary. This method does not introduce a three-dimensional terrain structure, cannot perceive slope / undulation, and is prone to boundary jumping or fragmentation. The cyclic residual convolution structure lacks a fusion mechanism for physical features (such as slope, sun altitude, etc.), has poor generalization ability under different terrain and remote sensing conditions, and is prone to overfitting small sample scenes.

[0004] However, traditional methods often rely on a single remote sensing image or use only two-dimensional images for boundary extraction, making them difficult to address in large-scale areas with complex terrain (such as mountain terraces), mixed vegetation, and broken boundaries. Furthermore, existing methods lack the ability to perceive terrain undulations and cannot effectively distinguish true plot boundaries from pseudo-edges caused by shadows, soil moisture, or reflective interference. With the advancement of remote sensing sensor capabilities, fusing multispectral remote sensing images, synthetic aperture radar (SAR) images, and digital elevation models (DEMs) has become an effective approach to improving boundary recognition accuracy. However, issues such as inconsistent resolution, complex spatial registration, and inconsistent features between multi-source data still pose significant challenges to fusion modeling. Therefore, a new method for farmland boundary recognition that fuses spectral, structural, and radar information is urgently needed to enhance boundary continuity, stability, and terrain rationality. Summary of the Invention

[0005] In view of this, the present invention provides a large-scale farmland boundary identification method and system based on multi-source remote sensing data fusion. A three-dimensional terrain structure model is constructed based on multispectral images, SAR images and digital elevation models. Spectral index and SAR polarization features are extracted and fused to generate a boundary probability map. By extracting coarse-grained farmland boundaries and introducing slope information to optimize the curve shape, fine boundary identification in complex terrain areas can be achieved.

[0006] To achieve the above-mentioned purpose, the present invention provides a large-scale farmland boundary recognition method using multi-source remote sensing data fusion, comprising the following steps:

[0007] S1: Collect multi-source remote sensing data of the farmland area, wherein the multi-source remote sensing data includes multispectral remote sensing images, SAR images, and digital elevation models, and use the digital elevation models of the farmland area to perform terrain perception on the farmland area and generate a three-dimensional terrain structure model of the farmland area;

[0008] S2: Adaptively perform spectral fusion on the multispectral remote sensing image to obtain a spectral fusion remote sensing image. Based on the collinearity equation correction principle, the pixel values ​​of the pixels in the spectral fusion remote sensing image are mapped to the three-dimensional coordinate points on the surface of the three-dimensional terrain structure model.

[0009] S3: extracting the SAR polarization feature image and the spectral index fusion image of the mapped three-dimensional terrain structure model surface, performing multimodal gated fusion on the extracted images, and obtaining a boundary probability image of the three-dimensional terrain structure model surface;

[0010] S4: Extract coarse-grained farmland boundaries from the boundary probability image and extract the slope image from the 3D terrain structure model. Combined with the slope information of the pixels in the slope image, constrained optimization is performed on the coarse-grained farmland boundaries to obtain smooth farmland boundary recognition results.

[0011] As a further improvement method of the present invention:

[0012] Optionally, multi-source remote sensing data of the farmland area is collected, and a digital elevation model of the farmland area is used to perform terrain perception on the farmland area to generate a three-dimensional terrain structure model of the farmland area, including:

[0013] Extracting a digital elevation model of the farmland area, wherein the digital elevation model records elevation values ​​of some two-dimensional geographic coordinates in the farmland area, and using the two-dimensional geographic coordinates and the corresponding elevation values ​​as point cloud data of the farmland area; specifically, the two-dimensional geographic coordinates are latitude and longitude coordinates;

[0014] The point cloud data is filtered using median filtering;

[0015] For each point cloud data, search for the nearest K neighbor point cloud data to form a local plane, and calculate the normal vector of the local plane as the normal vector of the point cloud data;

[0016] Extracting the normal vectors of all point cloud data to form a normal vector field, converting the normal vector field into the divergence value of the gradient field using a divergence calculation method, constructing the Poisson equation of the three-dimensional surface of the farmland area, and iteratively solving the Poisson equation using a conjugate gradient algorithm to obtain an implicit representation of the three-dimensional surface of the farmland area;

[0017] The three-dimensional surface of the farmland area obtained by solving the Poisson equation is converted into an explicit triangular mesh model by using a meshing method.

[0018] Based on the triangular mesh model, all two-dimensional geographic coordinates of the farmland area and their corresponding elevation values ​​are extracted to form a complete three-dimensional structural model of the farmland area. The three-dimensional structural model of the farmland area describes the elevation values ​​of the farmland area at different two-dimensional geographic coordinates, and the two-dimensional geographic coordinates and the corresponding elevation values ​​are used as three-dimensional coordinate points on the surface of the three-dimensional structural model.

[0019] Specifically, the SAR image is a synthetic aperture radar image, and the SAR radar is used to horizontally transmit horizontally polarized electromagnetic waves and vertically transmit vertically polarized electromagnetic waves to the two-dimensional geographic coordinate position of the farmland area, and receive echo signals horizontally and vertically respectively to obtain horizontally polarized echo signals, vertically polarized echo signals and cross-polarized echo signals of the two-dimensional geographic coordinate position. The horizontally polarized echo signal is an echo signal transmitted horizontally and received horizontally, the vertically polarized echo signal is an echo signal transmitted vertically and received vertically, the cross-polarized echo signal is an echo signal transmitted horizontally and received vertically, and the echo signal transmitted vertically and received horizontally. The intensity of the echo signal is calculated to form a scattering matrix of the two-dimensional geographic coordinate position, and the two-dimensional geographic coordinate position is mapped to the three-dimensional coordinate point on the surface of the three-dimensional structure model of the farmland area, and the scattering matrix is ​​used as the pixel value of the three-dimensional coordinate point in the SAR image.

[0020] Optionally, performing adaptive spectral fusion on the multispectral remote sensing image to obtain a spectral fused remote sensing image includes:

[0021] The multispectral remote sensing image is composed of spectral remote sensing images of multiple bands, and the bands include B3 band, B4 band, B8 band and B11 band; specifically, the B3 band, B4 band, B8 band and B11 band belong to the green light band, the red light band, the near infrared band and the short infrared band respectively;

[0022] Adaptive spatial attention calculation is performed on spectral remote sensing images of different bands to obtain the spatial attention weights of pixels in the spectral remote sensing images. The pixel values ​​of the pixels in the spectral remote sensing images are weighted to obtain the weighted pixel values ​​of the pixels in the multispectral remote sensing images in different bands. The vegetation index, water index and soil moisture index of different pixels in the multispectral remote sensing images are calculated. The vegetation index, water index and soil moisture index are used as the pixel values ​​of the pixels in the multispectral remote sensing images to construct a spectral fusion remote sensing image. Specifically, the spectral remote sensing image is a pixel matrix S with A1 rows and A2 columns. The pixel value of the pixel in the spectral remote sensing image is the reflectance of the pixel in the corresponding band. The adaptive spatial attention calculation method is:

[0023] Att S =δ(Conv 7×7 (Concat(AvgPool(S),MaxPool(S))));

[0024] Among them, Att S Represents the spatial attention weight matrix corresponding to the pixel matrix S, AvgPool((S)) represents the average pooling processing of the pixel matrix S, and MaxPool(S) represents the maximum pooling processing of the pixel matrix S. The average pooling processing result reflects the background smoothing characteristics of the pixel matrix, and the maximum pooling processing result reflects the edge and high-frequency pixel area of ​​the pixel matrix;

[0025] Concat(AvgPool(S), MaxPool(S)) represents the concatenation of the average pooling result AvgPool(S) and the maximum pooling result MaxPool((S)) according to the channel dimension. The Concat(AvgPool(S), MaxPool((S)) is a 2-channel feature map in the form of A1 rows and A2 columns.

[0026] Conv 7×7 (·) represents a 2D convolution operation with a convolution kernel size of 7×7, δ(·) represents the Sigmoid activation function, and the Att S is the spatial attention weight matrix of row A1 and column A2, the spatial attention weight matrix Att S Any element in is between 0 and 1, corresponding to the spatial attention weight of the pixel in the pixel matrix S;

[0027] The calculation formulas for the vegetation index, water index and soil moisture index of different pixels in the multispectral remote sensing image are as follows:

[0028]

[0029] Among them, S1(i,j), S2(i,j), S3(i,j) represent the vegetation index, water index and soil moisture index of the pixel in the i-th row and j-th column of the multispectral remote sensing image respectively, B3(i,j), B4(i,j), B8(i,j), B11(i,j) represent the weighted pixel values ​​of the pixel in the i-th row and j-th column of the spectral remote sensing image in the B3 band, B4 band, B8 band and B11 band respectively, i∈[1,A1], j∈[1,A2];

[0030] Specifically, healthy vegetation in farmland areas has stronger reflection in the near-infrared band, while withered vegetation or bare soil has stronger reflection in the red light band. Therefore, the vegetation index is calculated using the B4 and B8 bands to accurately identify crop-growing areas in farmland and surrounding unplanted areas. Water bodies have weaker reflection in the near-infrared band and stronger reflection in the green light band. Therefore, the water body index is calculated using the B3 and B8 bands to identify irrigation water bodies, moist areas and dry soil, helping to demarcate the boundary between farmland and water bodies. The short-wave infrared band is very sensitive to water absorption. Moist soil has weaker reflection in this band, while dry soil has stronger reflection. The difference in reflection between moist and dry soil in the near-infrared band is small, but the reflection of vegetation is stronger. The B11 and B8 bands are combined to calculate the soil moisture index as the degree of soil moisture, monitor the degree of soil moisture, and distinguish moist farmland from arid areas.

[0031] Before calculating each index, the spatial attention weights of pixels in different bands are calculated using the spatial attention method. According to the characteristics of different regions, the weights of each pixel in different bands are dynamically adjusted. The weights reflect the relative importance of different regions in the image, thereby improving the expression ability of the image spatial information. As a result, when calculating the vegetation index, water index and soil moisture index, the accuracy of important areas in the image (such as farmland, water bodies or wetlands) is improved, thereby improving the accuracy of farmland boundary recognition.

[0032] Optionally, mapping pixel values ​​of pixels in the spectrally fused remote sensing image to three-dimensional coordinate points on the surface of a three-dimensional terrain structure model includes:

[0033] Specifically, the collinearity equation correction principle is: a certain three-dimensional coordinate point in space is collinear with its pixel point on the sensor imaging plane and the three-dimensional coordinate point of the spatial imaging center, satisfying the projective geometric relationship;

[0034] Normalizing the three-dimensional coordinate points on the surface of the three-dimensional terrain structure model to obtain normalized three-dimensional coordinate points;

[0035] The normalized three-dimensional coordinate points are mapped using the RPC model to obtain the pixel coordinates of the normalized three-dimensional coordinate points in the spectrally fused remote sensing image, and the pixel values ​​of the pixel coordinates in the spectrally fused remote sensing image are extracted as the spectral index fusion values ​​of the three-dimensional coordinate points on the surface of the three-dimensional terrain structure model associated with the normalized three-dimensional coordinate points.

[0036] Specifically, the RPC model is used to describe the imaging geometric relationship of remote sensing images in the form of rational polynomials and is provided by suppliers of multi-source remote sensing data;

[0037] As a preferred algorithm of the present invention, the shadow compensation coefficient of the three-dimensional coordinate point on the surface of the three-dimensional terrain structure model is generated by combining the solar altitude angle at the time of multi-source remote sensing data acquisition and the scattering matrix of the three-dimensional coordinate point in the SAR image. The spectral index fusion value is weightedly compensated. The shadow compensation coefficient of the three-dimensional coordinate point ((x, y, z)) is:

[0038] in, represents the shadow compensation coefficient of the three-dimensional coordinate point (x, y, z) on the surface of the three-dimensional terrain structure model, λ(x, y, z) represents the adaptive weight, specifically, λ(x, y, z) = 0.2-0.2·min{0.5,θ(x, y, z) / θ th )},θ th represents the preset incident angle threshold, θ(x,y,z) represents the emission angle of the horizontally polarized electromagnetic wave emitted by the SAR to the two-dimensional geographic coordinate position corresponding to the three-dimensional coordinate point (x,y,z), (x * ,y * ,z * ) represents the location of the SAR, HV(x,y,z) represents the horizontal and vertical cross divergence in the scatter matrix of the three-dimensional coordinate point (x,y,z), and HH((x,y,z)) represents the horizontal divergence in the scatter matrix of the three-dimensional coordinate point (x,y,z);

[0039] θ sun Indicates the solar altitude angle at the time of multi-source remote sensing data collection;

[0040] Due to terrain obstruction, some farmland areas receive less solar radiation, resulting in local shadows or insufficient irradiation. The spectral values ​​in these areas will be artificially low, causing the index calculation results to underestimate the vegetation, water or humidity levels. Therefore, combined with the HV / HH ratio in the SAR image, the scattering response of the ground objects to different polarization waves can be estimated, revealing the "obscured" vegetation or structural signals, and compensating the spectral index of the three-dimensional coordinate points. Among them, HV is usually more sensitive to vegetation, and HH is sensitive to water and terrain changes. Correct the difference in solar radiation caused by the tilt of the terrain, enhance the surface reflection consistent with the direction of light, compensate for the attenuation of the backlit surface, make the boundaries of the farmland areas close to the hillside and the shade of the mountain clearer, which is conducive to the extraction of the farmland boundary, and use adaptive weights to control the size of the coefficient to avoid the shadow compensation coefficient being too large.

[0041] Optionally, extracting a SAR polarization feature image and a spectral index fusion image of the mapped three-dimensional terrain structure model surface includes:

[0042] The vertical top-down projection method is used to project the three-dimensional coordinate points on the surface of the mapped three-dimensional terrain structure model along the Z-axis to generate an index image in the two-dimensional plane view. The SAR polarization characteristics and spectral index of the three-dimensional coordinate points are extracted and used as the pixel values ​​of the projected pixels in the index image respectively. The SAR polarization characteristic image and spectral index fusion image of the mapped three-dimensional terrain structure model surface are constructed.

[0043] Specifically, the polarization characteristics of the three-dimensional coordinate point (x, y, z) are HV(x, y, z) / HH(x, y, z), θ((x, y, z), θ(x, y, z) is the emission angle of the horizontally polarized electromagnetic wave horizontally emitted by the SAR to the two-dimensional geographic coordinate position corresponding to the three-dimensional coordinate point (x, y, z), and the spectral index of the three-dimensional coordinate point (x, y, z) is S1((x, y, z), S2((x, y, z), S3((x, y, z),

[0044] S1(x,y,z), S2(x,y,z), S3(x,y,z) are the vegetation index, water index and soil moisture index of the three-dimensional coordinate point ((x,y,z)) respectively;

[0045] The pixel position of the three-dimensional coordinate point (x, y, z) in the index image is ((x, y).

[0046] Optionally, performing multimodal gated fusion on the extracted images to obtain a boundary probability image of the surface of the three-dimensional terrain structure model includes:

[0047] Calculate the fusion gating factor of the SAR polarization feature image and the spectral index fusion image. The calculation formula of the fusion gating factor is:

[0048] g=δ(W SAR I SAR +W Spec I Spec );

[0049] Among them, g represents the fusion gating factor, I SAR Represents the SAR polarization characteristic image, I Spec represents the spectral index fusion image, W SARrepresents the SAR weight matrix, W Spec represents the spectral index weight matrix, δ(·) represents the Sigmoid activation function;

[0050] The fusion gating factor is used to perform fusion calculation on the SAR polarization feature image and the spectral index fusion image to obtain a fusion image I:

[0051] I=g⊙tanh(W SAR I SAR )+(Ug)⊙tanh(W Spec I Spec );

[0052] Where U represents the unit element matrix whose elements are all 1 and whose scale is consistent with the index image, and ⊙ represents element-by-element multiplication;

[0053] The fused image I is input into the convolution classifier to generate the boundary probability of each pixel in the fused image I. The boundary probability is used as the pixel value to construct the edge probability image.

[0054] Specifically, the fusion gating factor includes a learnable dynamic weighting coefficient for each pixel. The fusion gating factor is in the form of a matrix with A rows and B columns, where A represents the number of three-dimensional coordinate points of the three-dimensional terrain structure model in the east-west direction, and B represents the number of three-dimensional coordinate points of the three-dimensional terrain structure model in the north-south direction. Each matrix element in the fusion gating factor is a dynamic weighting coefficient of the pixel position associated with the matrix element position. For each pixel position, it is judged based on the dynamic weighting coefficient whether it relies more on SAR polarization information (structure, texture) or spectral index information (vegetation, water body, soil). The closer the dynamic weighting coefficient is to 1, the more the pixel position is dominated by SAR features, and the pixel is more likely to be located in the boundary or texture mutation area. The closer the dynamic weighting coefficient is to 0, the more the pixel position is dominated by spectral index, and the more likely it is to be located in the vegetation or water area.

[0055] Optionally, extracting coarse-grained farmland boundaries from the boundary probability image includes:

[0056] Setting a boundary probability threshold, converting the boundary probability image into a binary image, performing a morphological dilation-erosion operation on the binary image, and removing isolated pixels and small broken segments;

[0057] Connected region analysis is performed on the binary image after morphological operation. All interconnected boundary regions are marked and connected to obtain multiple connected regions. The area of ​​each connected region is calculated, and the connected region with the largest area is selected as the candidate region of coarse-grained farmland boundary. The edge point sequence in the candidate region of coarse-grained farmland boundary is extracted using the contour tracing algorithm to form a closed curve, which is used as the coarse-grained farmland boundary.

[0058] Specifically, the pixel value of a pixel in the boundary probability map reflects the probability that the location is a farmland boundary. Through thresholding operations, low-probability noise is screened out, and only areas with clear structures and clear boundaries are retained. The broken edge segments are connected through dilation operations, and local noise or isolated response points are removed through erosion, thereby enhancing the structural integrity and topological connectivity of the boundary. Geometric enhancement modeling is performed on continuous structures such as farmland ridges, and the area of ​​the connected area is calculated, and the main area or the area larger than the area threshold is retained, effectively avoiding the interference of small-area abnormal boundaries on the final recognition results.

[0059] Optionally, extracting a slope image from the three-dimensional terrain structure model and combining the slope information of pixels in the slope image to perform constrained optimization on the coarse-grained farmland boundary includes:

[0060] Calculating the slope of a three-dimensional coordinate point in the three-dimensional terrain structure model and mapping the slope of the three-dimensional coordinate point to an index image to form a slope image, wherein the index image is a two-dimensional plan view of the three-dimensional terrain structure model, and the image size of the slope image is consistent with the boundary probability image;

[0061] A farmland boundary energy function based on slope constraints is constructed. The farmland boundary energy function takes the farmland boundary as input and outputs the farmland boundary energy. The farmland boundary energy includes internal energy that characterizes the smoothness and continuity of the farmland boundary, boundary image energy that drives the farmland boundary to a location with high boundary probability, and slope energy that penalizes crossing steep slopes.

[0062] The coarse-grained farmland boundary is used as the input of the farmland boundary energy function, and the gradient descent algorithm is used to evolve the coarse-grained farmland boundary until the boundary evolution result meets the convergence condition. The current boundary evolution result is output as a smoothed farmland boundary recognition result, and the farmland boundary recognition result is mapped to a two-dimensional plane view of the three-dimensional terrain structure model.

[0063] Specifically, the expression of the coarse-grained farmland boundary Q0 extracted from the boundary probability image is:

[0064]

[0065] Where Q0(r) represents the pixel position coordinates of the rth segment position in the coarse-grained farmland boundary Q0 in the boundary probability image. r∈[0,1], r represents the parameter for normalizing the length of farmland boundary;

[0066] The expression of the farmland boundary energy function is:

[0067] E(Q t )=w1·E1((Q t )+w2·E2(Q t,P)+w3·E2(Q t ,β);

[0068]

[0069] Among them, Q t represents the t-th boundary evolution result of the coarse-grained farmland boundary, E(·) represents the farmland boundary energy function, E(Q t ) represents Q t The farmland boundary energy, E1(Q t ) represents Q t The internal energy, E2((Q t ,P) represents Q t The boundary image energy, E2(Q t ,β) represents Q t Slope energy, P represents the boundary probability image, β represents the slope image; w1, w2, w3 represent the weights of internal energy, boundary image energy and slope energy respectively; the boundary image energy is preceded by a negative sign, which means maximizing the boundary probability, that is, encouraging the boundary to be close to the boundary probability strong response area;

[0070] Q t (r) represents Q t The pixel position coordinates of the rth segment in the boundary probability image, Represents Q t The first-order term of (r), Represents Q t The second-order term of (r);

[0071] The first-order term is an elastic term, representing the sum of the squared lengths of the line segments between boundary points, corresponding to the degree of stretching or compression of the boundary. When this term is small, it encourages equal spacing and avoids excessive dispersion between points, which is conducive to forming a compact structure and a well-closed boundary. The second-order term is a smoothing term, reflecting the local curvature of each boundary point relative to its neighbors, encouraging geometric continuity and smoothness of the boundary, and helping to avoid high-frequency oscillations in noisy areas.

[0072] P(Q t (r)) represents the pixel position coordinate Q in the boundary probability image t (Boundary probability at (r);

[0073] β(Q t (r)) represents the pixel position coordinate Q in the slope image t The slope at (r);

[0074] τ represents the slope suppression coefficient, represents the activation factor, if β(Q t (r)) is greater than 25 degrees, then otherwise dr represents the differential representation of parameter r.

[0075] In order to solve the above problems, the present invention provides a large-scale farmland boundary recognition system, which includes a server and a data acquisition device. The server includes a multimodal gated fusion module and a farmland boundary recognition module:

[0076] The multimodal gated fusion module is used to extract the SAR polarization feature image and the spectral index fusion image of the mapped three-dimensional terrain structure model surface, and perform multimodal gated fusion on the extracted images to obtain a boundary probability image of the three-dimensional terrain structure model surface;

[0077] The farmland boundary recognition module is used to extract coarse-grained farmland boundaries from the boundary probability image, and extract the slope image in the three-dimensional terrain structure model. Combined with the slope information of the pixels in the slope image, the coarse-grained farmland boundaries are constrained and optimized to obtain a smooth farmland boundary recognition result.

[0078] The data acquisition device is used to collect multi-source remote sensing data of a farmland area, use a digital elevation model of the farmland area to perform terrain perception on the farmland area, generate a three-dimensional terrain structure model of the farmland area, perform adaptive spectral fusion on the multispectral remote sensing image to obtain a spectral fusion remote sensing image, and map the pixel values ​​of pixels in the spectral fusion remote sensing image to three-dimensional coordinate points on the surface of the three-dimensional terrain structure model based on the collinearity equation correction principle;

[0079] To realize a large-scale farmland boundary recognition method based on any of the multi-source remote sensing data fusion methods mentioned above.

[0080] In order to solve the above problem, the present invention provides an electronic device, comprising:

[0081] a memory storing at least one instruction;

[0082] Communication interfaces to enable electronic equipment to communicate; and

[0083] The processor executes the instructions stored in the memory to implement the large-scale farmland boundary recognition method based on multi-source remote sensing data fusion.

[0084] In order to solve the above problems, the present invention also provides a computer-readable storage medium, which stores at least one instruction, and the at least one instruction is executed by a processor in an electronic device to implement the above-mentioned large-scale farmland boundary identification method based on multi-source remote sensing data fusion.

[0085] Compared with the existing technology, the present invention proposes a large-scale farmland boundary recognition method and system based on multi-source remote sensing data fusion. This technology has the following advantages:

[0086] First, the present application combines the solar altitude angle at the time of multi-source remote sensing data acquisition and the scattering matrix of the three-dimensional coordinate points in the SAR image to generate the shadow compensation coefficient of the three-dimensional coordinate points on the surface of the three-dimensional terrain structure model, and performs weighted compensation on the spectral index fusion value. Specifically, due to the existence of terrain obstruction, some farmland areas are difficult to receive sufficient solar radiation, which can easily form local shadows or insufficient irradiation, thereby causing the spectral value of the area in the remote sensing image to be low, causing parameters such as vegetation index, water index or soil moisture index to be underestimated. To overcome this problem, the present application fuses the polarization information in the SAR image, especially using the HV / HH polarization scattering ratio to estimate the response ability of the ground object to different polarization waves, thereby revealing the true reflection characteristics of the vegetation or surface structure that still exists in the shadow-covered area. Among them, HV polarization is more sensitive to vegetation structure and can supplement the vegetation signal lost due to shadow weakening; HH polarization is more sensitive to water body boundaries and undulating terrain, which helps to depict the reflection differences caused by terrain. At the same time, an adaptive weight mechanism is used to dynamically adjust the compensation coefficient to prevent the introduction of noise or artifacts due to over-compensation, ensuring that the compensation effect has physical consistency and regional adaptability.

[0087] At the same time, the farmland boundary curve energy function proposed in this application integrates the multimodal information of the boundary probability image and the slope image, and suppresses the boundary jump across the high slope area by introducing a gradient adjustment factor. At the same time, the geometric continuity control of the boundary is realized through the elastic term and the curvature term, which improves the physical rationality and spatial structure smoothness of the boundary curve as a whole. It is especially suitable for fine boundary extraction scenarios in areas with undulating terrain or multiple cultivated land structures. Specifically, by guiding the contour evolution by multiple factors, integrating internal energy (smoothness) + boundary guidance + slope suppression, spectral + terrain dual-mode constraints are realized to avoid false boundaries caused by spectral disturbances such as shadows and high reflective areas of the contour, especially for areas with strong undulations such as terraces, to solve the problems of "broken boundaries" and "folded boundaries", and introduce slope terms for constraints, corresponding to the terrain continuity hypothesis: the farmland boundary should not cross the sudden steep slope to avoid non-physical edges. By introducing first-order and second-order terms, and fusing them with the boundary probability image and the slope image, the contour can be adaptively adjusted to the strength of the image boundary. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] Figure 1 A flowchart of a large-scale farmland boundary recognition method using multi-source remote sensing data fusion is provided in accordance with an embodiment of the present invention.

[0089] Figure 2 A spectral index fusion image based on multi-source remote sensing data fusion is provided in one embodiment of the present invention.

[0090] Figure 3This is a comparison of the changes in vegetation index experimental indicators before and after shadow compensation provided by an embodiment of the present invention.

[0091] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0092] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0093] The embodiment of the present application provides a large-scale farmland boundary identification method based on multi-source remote sensing data fusion. The execution subject of the large-scale farmland boundary identification method based on multi-source remote sensing data fusion includes but is not limited to at least one of the electronic devices such as a server and a terminal that can be configured to execute the method provided by the embodiment of the present application. In other words, the large-scale farmland boundary identification method based on multi-source remote sensing data fusion can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0094] Reference Figure 1 , embodiment 1 of the present invention is:

[0095] A large-scale farmland boundary recognition method based on multi-source remote sensing data fusion includes the following steps:

[0096] S1: Collect multi-source remote sensing data of the farmland area, wherein the multi-source remote sensing data includes multispectral remote sensing images, SAR images and digital elevation models, use the digital elevation models of the farmland area to perform terrain perception on the farmland area, and generate a three-dimensional terrain structure model of the farmland area.

[0097] Collect multi-source remote sensing data of the farmland area, use the digital elevation model of the farmland area to perform terrain perception of the farmland area, and generate a three-dimensional terrain structure model of the farmland area, including:

[0098] Extracting a digital elevation model of the farmland area, wherein the digital elevation model records elevation values ​​of some two-dimensional geographic coordinates in the farmland area, and using the two-dimensional geographic coordinates and the corresponding elevation values ​​as point cloud data of the farmland area; specifically, the two-dimensional geographic coordinates are latitude and longitude coordinates;

[0099] The point cloud data is filtered using median filtering;

[0100] For each point cloud data, search for the nearest K neighbor point cloud data to form a local plane, and calculate the normal vector of the local plane as the normal vector of the point cloud data;

[0101] Extracting the normal vectors of all point cloud data to form a normal vector field, converting the normal vector field into the divergence value of the gradient field using a divergence calculation method, constructing the Poisson equation of the three-dimensional surface of the farmland area, and iteratively solving the Poisson equation using a conjugate gradient algorithm to obtain an implicit representation of the three-dimensional surface of the farmland area;

[0102] A meshing process is used to convert the implicitly represented three-dimensional surface of the farmland area obtained by solving the Poisson equation into an explicitly represented triangular mesh model; as an embodiment of the present invention, the meshing process uses a Delaunay triangulation algorithm or a Marching Cubes algorithm;

[0103] Based on the triangular mesh model, all two-dimensional geographic coordinates of the farmland area and their corresponding elevation values ​​are extracted to form a complete three-dimensional structural model of the farmland area. The three-dimensional structural model of the farmland area describes the elevation values ​​of the farmland area at different two-dimensional geographic coordinates, and the two-dimensional geographic coordinates and the corresponding elevation values ​​are used as three-dimensional coordinate points on the surface of the three-dimensional structural model.

[0104] S2: Adaptive spectral fusion is performed on the multispectral remote sensing image to obtain a spectral fused remote sensing image. Based on the collinearity equation correction principle, the pixel values ​​of the pixels in the spectral fused remote sensing image are mapped to the three-dimensional coordinate points on the surface of the three-dimensional terrain structure model.

[0105] Adaptive spectral fusion is performed on multispectral remote sensing images to obtain spectral fusion remote sensing images, including:

[0106] The multispectral remote sensing image is composed of spectral remote sensing images of multiple bands, and the bands include B3 band, B4 band, B8 band and B11 band; specifically, the B3 band, B4 band, B8 band and B11 band belong to the green light band, the red light band, the near infrared band and the short infrared band respectively;

[0107] Adaptive spatial attention calculation is performed on spectral remote sensing images of different bands to obtain the spatial attention weights of pixels in the spectral remote sensing images. The pixel values ​​of the pixels in the spectral remote sensing images are weighted to obtain the weighted pixel values ​​of the pixels in the multispectral remote sensing images in different bands. The vegetation index, water index and soil moisture index of different pixels in the multispectral remote sensing images are calculated. The vegetation index, water index and soil moisture index are used as the pixel values ​​of the pixels in the multispectral remote sensing images to construct a spectral fusion remote sensing image. Specifically, the spectral remote sensing image is a pixel matrix S with A1 rows and A2 columns. The pixel value of the pixel in the spectral remote sensing image is the reflectance of the pixel in the corresponding band. The adaptive spatial attention calculation method is:

[0108] Att S =δ(Conv 7×7(Concat(AvgPool(S),MaxPool(S))));

[0109] Among them, Att S Represents the spatial attention weight matrix corresponding to the pixel matrix S, AvgPool((S)) represents the average pooling processing of the pixel matrix S, and MaxPool(S) represents the maximum pooling processing of the pixel matrix S. The average pooling processing result reflects the background smoothing characteristics of the pixel matrix, and the maximum pooling processing result reflects the edge and high-frequency pixel area of ​​the pixel matrix;

[0110] Concat(AvgPool(S), MaxPool(S)) represents the concatenation of the average pooling result AvgPool(S) and the maximum pooling result MaxPool((S)) according to the channel dimension. The Concat(AvgPool(S), MaxPool((S)) is a 2-channel feature map in the form of A1 rows and A2 columns.

[0111] Conv 7×7 (·) represents a 2D convolution operation with a convolution kernel size of 7×7, δ(·) represents the Sigmoid activation function, and the Att S is the spatial attention weight matrix of row A1 and column A2, the spatial attention weight matrix Att S Any element in is between 0 and 1, corresponding to the spatial attention weight of the pixel in the pixel matrix S;

[0112] The calculation formulas for the vegetation index, water index and soil moisture index of different pixels in the multispectral remote sensing image are as follows:

[0113]

[0114] Among them, S1(i,j), S2(i,j), S3(i,j) represent the vegetation index, water index and soil moisture index of the pixel in the i-th row and j-th column of the multispectral remote sensing image respectively, B3(i,j), B4(i,j), B8(i,j), B11(i,j) represent the weighted pixel values ​​of the pixel in the i-th row and j-th column of the spectral remote sensing image in the B3 band, B4 band, B8 band and B11 band respectively, i∈[1,A1], j∈[1,A2].

[0115] Mapping pixel values ​​of pixels in the spectral fusion remote sensing image to three-dimensional coordinate points on the surface of a three-dimensional terrain structure model includes:

[0116] Normalizing the three-dimensional coordinate points on the surface of the three-dimensional terrain structure model to obtain normalized three-dimensional coordinate points;

[0117] The normalized three-dimensional coordinate points are mapped using the RPC model to obtain the pixel coordinates of the normalized three-dimensional coordinate points in the spectrally fused remote sensing image, and the pixel values ​​of the pixel coordinates in the spectrally fused remote sensing image are extracted as the spectral index fusion values ​​of the three-dimensional coordinate points on the surface of the three-dimensional terrain structure model associated with the normalized three-dimensional coordinate points.

[0118] As a preferred algorithm of the present invention, the shadow compensation coefficient of the three-dimensional coordinate point on the surface of the three-dimensional terrain structure model is generated by combining the solar altitude angle at the time of multi-source remote sensing data acquisition and the scattering matrix of the three-dimensional coordinate point in the SAR image. The spectral index fusion value is weightedly compensated. The shadow compensation coefficient of the three-dimensional coordinate point ((x, y, z)) is:

[0119]

[0120] in, represents the shadow compensation coefficient of the three-dimensional coordinate point (x, y, z) on the surface of the three-dimensional terrain structure model, λ(x, y, z) represents the adaptive weight, specifically, λ(x, y, z) = 0.2-0.2·min{0.5,θ(x, y, z) / θ th )},θ th represents the preset incident angle threshold, θ(x,y,z) represents the emission angle of the horizontally polarized electromagnetic wave emitted by the SAR to the two-dimensional geographic coordinate position corresponding to the three-dimensional coordinate point (x,y,z), (x * ,y * ,z * ) represents the location of the SAR, HV(x,y,z) represents the horizontal and vertical cross divergence in the scatter matrix of the three-dimensional coordinate point (x,y,z), and HH((x,y,z)) represents the horizontal divergence in the scatter matrix of the three-dimensional coordinate point (x,y,z);

[0121] θ sun Indicates the solar altitude angle at the time of multi-source remote sensing data collection;

[0122] As an embodiment of the present invention, if the pixel coordinate corresponding to the normalized three-dimensional coordinate point does not have a pixel value, a bilinear interpolation method is used in the spectrally fused remote sensing image to obtain the pixel value of the pixel coordinate.

[0123] S3: Extract the SAR polarization feature image and spectral index fusion image of the mapped three-dimensional terrain structure model surface, perform multimodal gated fusion on the extracted images, and obtain the boundary probability image of the three-dimensional terrain structure model surface.

[0124] Extract the SAR polarization feature image and spectral index fusion image of the mapped three-dimensional terrain structure model surface, including:

[0125] The vertical top-down projection method is used to project the three-dimensional coordinate points on the surface of the mapped three-dimensional terrain structure model along the Z-axis to generate an index image in the two-dimensional plane view. The SAR polarization characteristics and spectral index of the three-dimensional coordinate points are extracted and used as the pixel values ​​of the projected pixels in the index image respectively. The SAR polarization characteristic image and spectral index fusion image of the mapped three-dimensional terrain structure model surface are constructed.

[0126] Specifically, the polarization characteristics of the three-dimensional coordinate point (x, y, z) are HV(x, y, z) / HH(x, y, z), θ((x, y, z), θ(x, y, z) is the emission angle of the horizontally polarized electromagnetic wave horizontally emitted by the SAR to the two-dimensional geographic coordinate position corresponding to the three-dimensional coordinate point (x, y, z), and the spectral index of the three-dimensional coordinate point (x, y, z) is S1((x, y, z), S2((x, y, z), S3((x, y, z),

[0127] S1(x,y,z), S2(x,y,z), S3(x,y,z) are the vegetation index, water index and soil moisture index of the three-dimensional coordinate point ((x,y,z)) respectively;

[0128] The pixel position of the three-dimensional coordinate point (x, y, z) in the index image is ((x, y).

[0129] The extracted images are subjected to multimodal gated fusion to obtain a boundary probability image of the surface of the 3D terrain structure model, including:

[0130] Calculate the fusion gating factor of the SAR polarization feature image and the spectral index fusion image. The calculation formula of the fusion gating factor is:

[0131] g=δ(W SAR I SAR +W Spec I Spec );

[0132] Among them, g represents the fusion gating factor, I SAR Represents the SAR polarization characteristic image, I Spec represents the spectral index fusion image, W SAR represents the SAR weight matrix, W Spec represents the spectral index weight matrix, δ(·) represents the Sigmoid activation function;

[0133] The fusion gating factor is used to perform fusion calculation on the SAR polarization feature image and the spectral index fusion image to obtain a fusion image I:

[0134] I=g⊙tanh(W SAR I SAR)+(Ug)⊙tanh(W Spec I Spec );

[0135] Where U represents the unit element matrix whose elements are all 1 and whose scale is consistent with the index image, and ⊙ represents element-by-element multiplication;

[0136] The fused image I is input into the convolution classifier to generate the boundary probability of each pixel in the fused image I. The boundary probability is used as the pixel value to construct the edge probability image.

[0137] As an embodiment of the present invention, the index image is in the form of a pixel matrix with A rows and B columns, and the SAR polarization characteristic image I SAR It is a multi-channel pixel matrix with 2 channels, A rows and B columns. The SAR polarization feature image I SAR The pixel value of the pixel in is the two-dimensional polarization feature, and the spectral index fusion image I Spec It is a multi-channel pixel matrix with 3 channels in rows A and columns B, where the 3 channels correspond to 3 spectral indices, W SAR It is a 2-channel A-row B-column weight matrix, W Spec For a 3-channel weight matrix with A rows and B columns, then W SAR I SAR ,W Spec I Spec It is a single-channel pixel matrix with A rows and B columns, where A represents the number of three-dimensional coordinate points of the three-dimensional terrain structure model in the east-west direction, and B represents the number of three-dimensional coordinate points of the three-dimensional terrain structure model in the north-south direction.

[0138] S4: Extract coarse-grained farmland boundaries from the boundary probability image and extract the slope image from the 3D terrain structure model. Combined with the slope information of the pixels in the slope image, constrained optimization is performed on the coarse-grained farmland boundaries to obtain smooth farmland boundary recognition results.

[0139] Extract coarse-grained farmland boundaries from the boundary probability image, including:

[0140] Setting a boundary probability threshold, converting the boundary probability image into a binary image, performing a morphological dilation-erosion operation on the binary image, and removing isolated pixels and small broken segments;

[0141] Connected region analysis is performed on the binary image after morphological operation. All interconnected boundary regions are marked and connected to obtain multiple connected regions. The area of ​​each connected region is calculated, and the connected region with the largest area is selected as the candidate region of coarse-grained farmland boundary. The edge point sequence in the candidate region of coarse-grained farmland boundary is extracted using the contour tracing algorithm to form a closed curve, which is used as the coarse-grained farmland boundary.

[0142] As an embodiment of the present invention, if there are multiple field structures in the farmland area, connected areas with an area greater than a preset area threshold are screened, and the coarse-grained farmland boundaries of each connected area are extracted respectively. The contour tracking algorithm is the Suzuki boundary tracking algorithm.

[0143] Extracting a slope image from a three-dimensional terrain structure model, combining the slope information of pixels in the slope image, and performing constrained optimization on the coarse-grained farmland boundary, including:

[0144] Calculating the slope of a three-dimensional coordinate point in the three-dimensional terrain structure model and mapping the slope of the three-dimensional coordinate point to an index image to form a slope image, wherein the index image is a two-dimensional plan view of the three-dimensional terrain structure model, and the image size of the slope image is consistent with the boundary probability image;

[0145] In an embodiment of the present invention, the slope of the three-dimensional coordinate point (x, y, z) is:

[0146]

[0147] Where β(x,y,z) represents the slope of the three-dimensional coordinate point (x,y,z), Δ1 represents the coordinate point width of adjacent three-dimensional coordinate points in the longitude direction in the three-dimensional terrain structure model, Δ2 represents the coordinate point width of adjacent three-dimensional coordinate points in the latitude direction in the three-dimensional terrain structure model, and Z(x+1,y) represents the elevation value of the two-dimensional geographic coordinate (x+1,y) in the three-dimensional terrain structure model.

[0148] The pixel coordinates of the three-dimensional coordinate point (x, y, z) in the slope image are ((x, y), and the pixel value is β(x, y, z);

[0149] A farmland boundary energy function based on slope constraints is constructed. The farmland boundary energy function takes the farmland boundary as input and outputs the farmland boundary energy. The farmland boundary energy includes internal energy that characterizes the smoothness and continuity of the farmland boundary, boundary image energy that drives the farmland boundary to a location with high boundary probability, and slope energy that penalizes crossing steep slopes.

[0150] The coarse-grained farmland boundary is used as input to a farmland boundary energy function. A gradient descent algorithm is used to evolve the coarse-grained farmland boundary until the boundary evolution result satisfies a convergence condition. The current boundary evolution result is output as a smoothed farmland boundary identification result, and the farmland boundary identification result is mapped to a two-dimensional plan view of a three-dimensional terrain structure model. In one embodiment of the present invention, the convergence condition is that the change in the farmland boundary energy is less than a preset energy change threshold.

[0151] Specifically, the expression of the coarse-grained farmland boundary Q0 extracted from the boundary probability image is:

[0152]

[0153] Where Q0(r) represents the pixel position coordinates of the rth segment position in the coarse-grained farmland boundary Q0 in the boundary probability image. r represents the parameter for normalizing the length of the farmland boundary;

[0154] The expression of the farmland boundary energy function is:

[0155] E(Q t )=w1·E1((Q t )+w2·E2(Q t ,P)+w3·E2(Q t ,β);

[0156]

[0157] Among them, Q t represents the t-th boundary evolution result of the coarse-grained farmland boundary, E(·) represents the farmland boundary energy function, E(Q t ) represents Q t The farmland boundary energy, E1(Q t ) represents Q t The internal energy, E2((Q t ,P) represents Q t The boundary image energy, E2(Q t ,β) represents Q t Slope energy, P represents the boundary probability image, β represents the slope image; w1, w2, w3 represent the weights of internal energy, boundary image energy and slope energy respectively; the boundary image energy is preceded by a negative sign, which means maximizing the boundary probability, that is, encouraging the boundary to be close to the boundary probability strong response area;

[0158] Q t (r) represents Q t The pixel position coordinates of the rth segment in the boundary probability image, Represents Q t The first-order term of (r), Represents Q t The second-order term of (r);

[0159] The first-order term is an elastic term, representing the sum of the squared lengths of the line segments between boundary points, corresponding to the degree of stretching or compression of the boundary. When this term is small, it encourages equal spacing and avoids excessive dispersion between points, which is conducive to forming a compact structure and a well-closed boundary. The second-order term is a smoothing term, reflecting the local curvature of each boundary point relative to its neighbors, encouraging geometric continuity and smoothness of the boundary, and helping to avoid high-frequency oscillations in noisy areas.

[0160] P(Q t (r)) represents the pixel position coordinate Q in the boundary probability image t (Boundary probability at (r);

[0161] β(Q t (r)) represents the pixel position coordinate Q in the slope image t The slope at (r);

[0162] τ represents the slope suppression coefficient, represents the activation factor, if β(Q t (r)) is greater than 25 degrees, then otherwise dr represents the differential representation of parameter r.

[0163] Example 2:

[0164] like Figure 2 The spectral index fusion image based on multi-source remote sensing data fusion in the case of multiple farmlands is shown.

[0165] Example 3:

[0166] Taking vegetation index as an example, Figure 3 The figure shows the comparative changes of vegetation index experimental indicators before and after shadow compensation. The experimental area is a typical mountainous and hilly farmland area. The traditional uncompensated vegetation index method is compared with the vegetation index calculation method proposed in this invention that introduces the shadow compensation coefficient. The comparison results show that:

[0167] 1. In the shadowed area of ​​the slope, the local variance of the traditional vegetation index is 0.041, but after compensation using the proposed method, it is reduced to 0.017, indicating that the vegetation index is smoother and more stable in areas with uneven illumination.

[0168] 2. The mean boundary gradient of the traditional vegetation index is 0.11, but it is increased to 0.22 using the compensation method, which enhances the contrast of the boundary of the ground objects and facilitates the identification of farmland boundaries.

[0169] 3. Compared with the ground-measured vegetation index, the root mean square error between the traditional vegetation index and the ground-measured vegetation index is 0.082, while the root mean square error after compensation of the present invention is 0.046, showing higher field consistency.

[0170] Example 4:

[0171] This embodiment provides a large-scale farmland boundary recognition system, which includes a server and a data acquisition device. The server includes a multimodal gated fusion module and a farmland boundary recognition module.

[0172] The multimodal gated fusion module is used to extract the SAR polarization feature image and the spectral index fusion image of the mapped three-dimensional terrain structure model surface, and perform multimodal gated fusion on the extracted images to obtain a boundary probability image of the three-dimensional terrain structure model surface;

[0173] The farmland boundary recognition module is used to extract coarse-grained farmland boundaries from the boundary probability image, and extract the slope image in the three-dimensional terrain structure model. Combined with the slope information of the pixels in the slope image, the coarse-grained farmland boundaries are constrained and optimized to obtain a smooth farmland boundary recognition result.

[0174] The data acquisition device is used to collect multi-source remote sensing data of a farmland area, use a digital elevation model of the farmland area to perform terrain perception on the farmland area, generate a three-dimensional terrain structure model of the farmland area, perform adaptive spectral fusion on the multispectral remote sensing image to obtain a spectral fusion remote sensing image, and map the pixel values ​​of pixels in the spectral fusion remote sensing image to three-dimensional coordinate points on the surface of the three-dimensional terrain structure model based on the collinearity equation correction principle;

[0175] This is to implement the large-scale farmland boundary recognition method of multi-source remote sensing data fusion in Example 1.

[0176] It should be understood that the embodiment is for illustration only and the scope of the patent application is not limited to this structure.

[0177] It should be noted that the serial numbers of the above-mentioned embodiments of the present invention are for descriptive purposes only and do not represent the advantages or disadvantages of the embodiments. In addition, the terms "including", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method comprising the element.

[0178] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better embodiment. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0179] The above are only preferred embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A large-scale farmland boundary recognition method based on multi-source remote sensing data fusion, characterized by: The method comprises: S1: Collect multi-source remote sensing data of the farmland area, wherein the multi-source remote sensing data includes multispectral remote sensing images, SAR images, and digital elevation models, and use the digital elevation models of the farmland area to perform terrain perception on the farmland area and generate a three-dimensional terrain structure model of the farmland area; The pixel coordinates of the SAR image are the three-dimensional coordinate points on the surface of the three-dimensional structure model, and the pixel values ​​are the scattering matrix of the three-dimensional coordinate points; S2: Adaptively perform spectral fusion on the multispectral remote sensing image to obtain a spectral fusion remote sensing image. Based on the collinearity equation correction principle, the pixel values ​​of the pixels in the spectral fusion remote sensing image are mapped to the three-dimensional coordinate points on the surface of the three-dimensional terrain structure model. The pixel values ​​of the pixels in the spectral fusion remote sensing image include vegetation index, water body index and soil moisture index; S3: extracting the SAR polarization feature image and the spectral index fusion image of the mapped three-dimensional terrain structure model surface, performing multimodal gated fusion on the extracted images, and obtaining a boundary probability image of the three-dimensional terrain structure model surface, wherein the pixel value of a pixel in the boundary probability image is the probability that the pixel is a farmland boundary; S4: Extract coarse-grained farmland boundaries from the boundary probability image and the slope image in the 3D terrain structure model. Combined with the slope information of the pixels in the slope image, constrained optimization is performed on the coarse-grained farmland boundaries to obtain smooth farmland boundary recognition results.

2. The large-scale farmland boundary recognition method based on multi-source remote sensing data fusion according to claim 1 is characterized in that: Collect multi-source remote sensing data of the farmland area, use the digital elevation model of the farmland area to perform terrain perception of the farmland area, and generate a three-dimensional terrain structure model of the farmland area, including: Extracting a digital elevation model of the farmland area, wherein the digital elevation model records elevation values ​​of some two-dimensional geographic coordinates in the farmland area, and using the two-dimensional geographic coordinates and the corresponding elevation values ​​as point cloud data of the farmland area; The point cloud data is filtered using median filtering; For each point cloud data, search for the nearest K neighbor point cloud data to form a local plane, and calculate the normal vector of the local plane as the normal vector of the point cloud data; Extracting the normal vectors of all point cloud data to form a normal vector field, converting the normal vector field into the divergence value of the gradient field using a divergence calculation method, constructing the Poisson equation of the three-dimensional surface of the farmland area, and iteratively solving the Poisson equation using a conjugate gradient algorithm to obtain an implicit representation of the three-dimensional surface of the farmland area; The three-dimensional surface of the farmland area obtained by solving the Poisson equation is converted into an explicit triangular mesh model by using a meshing method. Based on the triangular mesh model, all two-dimensional geographic coordinates of the farmland area and their corresponding elevation values ​​are extracted to form a complete three-dimensional structural model of the farmland area. The three-dimensional structural model of the farmland area describes the elevation values ​​of the farmland area at different two-dimensional geographic coordinates, and the two-dimensional geographic coordinates and the corresponding elevation values ​​are used as three-dimensional coordinate points on the surface of the three-dimensional structural model.

3. The large-scale farmland boundary recognition method based on multi-source remote sensing data fusion according to claim 1, characterized in that: Adaptive spectral fusion is performed on multispectral remote sensing images to obtain spectral fusion remote sensing images, including: The multispectral remote sensing image is composed of spectral remote sensing images of multiple bands; Adaptive spatial attention calculation is performed on spectral remote sensing images of different bands to obtain spatial attention weights of pixels in the spectral remote sensing images, pixel values ​​of pixels in the spectral remote sensing images are weighted to obtain weighted pixel values ​​of pixels in different bands in the multispectral remote sensing image, vegetation index, water index and soil moisture index of different pixels in the multispectral remote sensing image are calculated, and the vegetation index, water index and soil moisture index are used as pixel values ​​of pixels in the multispectral remote sensing image to construct a spectral fusion remote sensing image.

4. The large-scale farmland boundary recognition method based on multi-source remote sensing data fusion according to claim 3 is characterized in that: Mapping pixel values ​​of pixels in the spectral fusion remote sensing image to three-dimensional coordinate points on the surface of a three-dimensional terrain structure model includes: Normalizing the three-dimensional coordinate points on the surface of the three-dimensional terrain structure model to obtain normalized three-dimensional coordinate points; The normalized three-dimensional coordinate points are mapped using the RPC model to obtain the pixel coordinates of the normalized three-dimensional coordinate points in the spectrally fused remote sensing image, and the pixel values ​​of the pixel coordinates in the spectrally fused remote sensing image are extracted as the spectral index fusion values ​​of the three-dimensional coordinate points on the surface of the three-dimensional terrain structure model associated with the normalized three-dimensional coordinate points.

5. The large-scale farmland boundary recognition method based on multi-source remote sensing data fusion according to claim 1, characterized in that: Extract the SAR polarization feature image and spectral index fusion image of the mapped three-dimensional terrain structure model surface, including: The vertical top-down projection method is used to project the three-dimensional coordinate points on the surface of the mapped three-dimensional terrain structure model along the Z-axis to generate an index image in the two-dimensional plane view. The SAR polarization characteristics and spectral index of the three-dimensional coordinate points are extracted and used as the pixel values ​​of the projected pixels in the index image respectively. The SAR polarization characteristic image and spectral index fusion image of the mapped three-dimensional terrain structure model surface are constructed.

6. The large-scale farmland boundary recognition method based on multi-source remote sensing data fusion according to claim 5, characterized in that: The extracted images are subjected to multimodal gated fusion to obtain a boundary probability image of the surface of the 3D terrain structure model, including: Calculate the fusion gating factor of the SAR polarization feature image and the spectral index fusion image. The calculation formula of the fusion gating factor is: g=δ(W SAR ·I SAR +W Spec ·I Spec ); Among them, g represents the fusion gating factor, I SAR Represents the SAR polarization characteristic image, I Spec represents the spectral index fusion image, W SAR represents the SAR weight matrix, W Spec represents the spectral index weight matrix, δ(·) represents the Sigmoid activation function; The fusion gating factor is used to perform fusion calculation on the SAR polarization feature image and the spectral index fusion image to obtain a fusion image I: I=g⊙tanh(W SAR ·I SAR )+(U-g)⊙tanh(W Spec ·I Spec ); Where U represents the unit element matrix whose elements are all 1 and whose scale is consistent with the index image, and ⊙ represents element-by-element multiplication; The fused image I is input into the convolution classifier to generate the boundary probability of each pixel in the fused image I. The boundary probability is used as the pixel value to construct the edge probability image.

7. The large-scale farmland boundary recognition method based on multi-source remote sensing data fusion according to claim 6, characterized in that: Extracting a coarse-grained farmland boundary from the boundary probability image includes: Setting a boundary probability threshold, converting the boundary probability image into a binary image, performing a morphological dilation-erosion operation on the binary image, and removing isolated pixels and small broken segments; Connected region analysis is performed on the binary image after morphological operation. All interconnected boundary regions are marked and connected to obtain multiple connected regions. The area of ​​each connected region is calculated, and the connected region with the largest area is selected as the candidate region of coarse-grained farmland boundary. The edge point sequence in the candidate region of coarse-grained farmland boundary is extracted using the contour tracing algorithm to form a closed curve, which is used as the coarse-grained farmland boundary.

8. The large-scale farmland boundary recognition method based on multi-source remote sensing data fusion according to claim 7, characterized in that: Extracting a slope image from a three-dimensional terrain structure model, combining the slope information of pixels in the slope image, and performing constrained optimization on the coarse-grained farmland boundary, including: Calculating the slope of a three-dimensional coordinate point in the three-dimensional terrain structure model and mapping the slope of the three-dimensional coordinate point to an index image to form a slope image, wherein the index image is a two-dimensional plan view of the three-dimensional terrain structure model, and the image size of the slope image is consistent with the boundary probability image; A farmland boundary energy function based on slope constraints is constructed. The farmland boundary energy function takes the farmland boundary as input and outputs the farmland boundary energy. The farmland boundary energy includes internal energy that characterizes the smoothness and continuity of the farmland boundary, boundary image energy that drives the farmland boundary to a location with high boundary probability, and slope energy that penalizes crossing steep slopes. The coarse-grained farmland boundary is used as the input of the farmland boundary energy function, and the gradient descent algorithm is used to evolve the coarse-grained farmland boundary until the boundary evolution result meets the convergence condition. The current boundary evolution result is output as a smoothed farmland boundary recognition result, and the farmland boundary recognition result is mapped to a two-dimensional plane view of the three-dimensional terrain structure model.

9. A large-scale farmland boundary recognition system, characterized by: The large-scale farmland boundary recognition system includes a server and a data acquisition device. The server includes a multimodal gated fusion module and a farmland boundary recognition module. The multimodal gated fusion module is used to extract the SAR polarization feature image and the spectral index fusion image of the mapped three-dimensional terrain structure model surface, and perform multimodal gated fusion on the extracted images to obtain a boundary probability image of the three-dimensional terrain structure model surface; The farmland boundary recognition module is used to extract coarse-grained farmland boundaries from the boundary probability image, and extract the slope image in the three-dimensional terrain structure model. Combined with the slope information of the pixels in the slope image, the coarse-grained farmland boundaries are constrained and optimized to obtain a smooth farmland boundary recognition result. The data acquisition device is used to collect multi-source remote sensing data of a farmland area, use a digital elevation model of the farmland area to perform terrain perception on the farmland area, generate a three-dimensional terrain structure model of the farmland area, perform adaptive spectral fusion on the multispectral remote sensing image to obtain a spectral fusion remote sensing image, and map the pixel values ​​of pixels in the spectral fusion remote sensing image to three-dimensional coordinate points on the surface of the three-dimensional terrain structure model based on the collinearity equation correction principle; To realize a large-scale farmland boundary recognition method of multi-source remote sensing data fusion as described in any one of claims 1-9.

Images