Forest land landslide collapse area detection method based on multi-view unmanned aerial vehicle remote sensing

Abstract

A forest land landslide collapse area detection method based on multi-view unmanned aerial vehicle remote sensing is characterized by comprising the following steps: step 1, constructing a forest land landslide collapse area detection system based on multi-view unmanned aerial vehicle remote sensing; 2, an air route planning module performs aerial photography route planning on the forest land target area; 3, the unmanned aerial vehicle aircraft carries out image shooting on the forest land target area to obtain an image data set; 4, performing sparse point cloud reconstruction on the image data set by the three-dimensional model construction module to obtain sparse point cloud; 5, performing high-precision three-dimensional reconstruction on the sparse point cloud by a three-dimensional model construction module to obtain a three-dimensional model; step 6, segmenting a forest land landslide collapse area in the three-dimensional model by an area identification calculation module, and determining forest land landslide area data; and 7, the area identification calculation module calculates the triangular area of each forest land landslide area and accumulates the triangular areas to obtain total area data. The method has the advantages that automatic recognition and area calculation of the forest land landslide collapse area are achieved.

Description

Technical Field

[0001] The present invention relates to the field of image processing information technology, and particularly to a method for detecting the area of forest land landslides and collapses based on multi-view UAV remote sensing. Background Art

[0002] With the intensification of climate change and human activities, mountain landslides and collapses occur frequently, causing serious impacts on the ecological environment and human life. As an important type of these disasters, forest land landslides and collapses have become one of the important fields of geological disaster research due to their concealment and destructiveness.

[0003] Accurate measurement of the landslide area is a key link in disaster assessment. It can not only intuitively reflect the scale and intensity of the landslide, but also provide an important basis for subsequent ecological restoration and geological treatment. Accurate landslide area data can help decision-makers better allocate resources, formulate targeted restoration strategies, and evaluate the restoration effect. In addition, time series analysis of landslide area data can reveal the development trend of landslides, providing scientific support for the establishment and improvement of early warning systems.

[0004] Disadvantages of the prior art: Traditional landslide and collapse detection methods usually rely on ground monitoring and satellite remote sensing data, but these methods have problems such as low resolution, poor timeliness, and inability to accurately monitor small areas. Summary of the Invention

[0005] A method for detecting the area of forest land landslides and collapses based on multi-view UAV remote sensing provided by the present invention realizes automatic identification and area calculation of forest land landslide and collapse areas, with high resolution, high detection accuracy, and good timeliness.

[0006] To achieve the above object, a key aspect of a method for detecting the area of forest land landslides and collapses based on multi-view UAV remote sensing provided by the present invention includes the following steps:

[0007] Step 1: Construct a system for detecting the area of forest land landslides and collapses based on multi-view UAV remote sensing. The system for detecting the area of forest land landslides and collapses includes a UAV aircraft, which is equipped with a GPS, a radar, and a high-definition camera. A route planning module, a three-dimensional model construction module, and a region identification and calculation module are also set in the UAV aircraft;

[0008] Step 2: The route planning module plans an aerial photography route for the image acquisition task of the forest land target area;

[0009] Step 3: The UAV aircraft takes images of the forest land target area according to the aerial photography route planned by the route planning module to obtain an image data set;

[0010] Step 4: The 3D model construction module uses the SFM method to perform sparse point cloud reconstruction on the image dataset to obtain the sparse point cloud of the woodland target area;

[0011] Step 5: The 3D model construction module uses the MVS method to perform high-precision 3D reconstruction on the sparse point cloud to obtain the 3D model of the refined target area;

[0012] Step 6: The area recognition and calculation module uses the Everything Segmentation SAM algorithm to segment the woodland landslide and collapse area in the 3D model to determine the woodland landslide area data;

[0013] Step 7: The area recognition and calculation module, based on the woodland landslide area data, obtains all the triangular elements used in the triangulation modeling for each woodland landslide area, then uses Heron's formula to calculate the area of each triangle in the woodland landslide area, and finally accumulates the areas of all the triangles in the woodland landslide areas to obtain the total area data of the woodland landslide areas.

[0014] Through the above design, first, based on the flight route planning method of five-way flight, the aerial photography route planning for the image acquisition task of the woodland landslide and collapse area is carried out to determine the UAV flight mode; then, the UAV equipped with GPS, radar, and high-definition camera is used to collect the target area images according to the planned route; then, the SFM method is used to reconstruct the sparse point cloud of the target area; then, the MVS method is used for refined 3D model reconstruction; then, based on the method of artificial intelligence everything segmentation and the 3D model image segmentation idea, the woodland landslide and collapse area is determined; finally, the triangular elements of the woodland landslide and collapse area are obtained, and the area of the triangular elements is calculated through Heron's formula to obtain the total area.

[0015] Through the high-resolution images taken by the multi-view UAVs and combined with computer vision and deep learning technologies, the accurate identification and area measurement of the landslide and collapse area are realized, which have the characteristics of high resolution, high detection accuracy, and good timeliness.

[0016] Preferably: In the step 2, the route planning module uses the five-way flight method for aerial photography route planning. The five-way flight method is specifically as follows: the route overlap rate is 85%, the side overlap rate is 80%, the orthophoto route shooting angle is 90 degrees, the oblique photography shooting angle is 45 degrees, and the flight mode uses terrain-following flight with a height difference of 15m. To ensure the fineness of the subsequent 3D reconstruction model.

[0017] According to the complexity and size of the target area terrain, the UAV uses the terrain-following flight height H; determine the appropriate flight speed v to ensure that there is enough time for the UAV to complete the exposure of the image between each shooting.

[0018] Preferably: in the step 4, the three-dimensional model construction module uses the SFM method to perform sparse point cloud reconstruction on the image data set, and the specific steps are as follows:

[0019] Step 41: The three-dimensional model construction module uses the SIFT algorithm to extract the feature points of each image data, find the key points of the image, and generate the descriptors related to each feature point; then uses the Fast Library for Approximate Nearest Neighbors (FLANN) to match the extracted feature points, obtain the matching point pairs, and calculate the Euclidean distance d between the descriptor vectors corresponding to each matching point pair; the approximate nearest neighbor algorithm can accelerate the matching between feature points.

[0020] The calculation formula for the Euclidean distance of the matching point pairs is as follows:

[0021]

[0022] where p i and q i are the descriptor vectors of the two feature points in the matching point pair;

[0023] Step 42: Based on the matching point pairs, the SFM algorithm is used to reconstruct the sparse point cloud, and the calculation formula is as follows:

[0024]

[0025] where P1 and P2 are the projection matrices of two cameras respectively, x1 and x2 are the pixel coordinates of the matching feature points in two images, and X p is the three-dimensional coordinate point in the sparse point cloud.

[0026] Preferably: in the step 5, the three-dimensional model construction module uses the MVS method to perform high-precision three-dimensional reconstruction on the sparse point cloud, and the specific steps are as follows:

[0027] Step 51: Based on the sparse point cloud, the three-dimensional model construction module uses the MVS method to generate a high-density point cloud, that is, by performing geometric inference on the multi-view image data set, the sparse point cloud is expanded into a dense point cloud;

[0028] The expression for point cloud reconstruction using the MVS method is:

[0029] P d = MVS(P s , I)

[0030] where I is the multi-view image data set, and P s ={p i |p i ∈R 3 , i = 1, 2,..., n s} is the input sparse point cloud, n s is the number of points in the sparse point cloud, P d ={q j |q j ∈R 3 , j = 1, 2, ..., n d} is the output dense point cloud, n d is the number of points in the dense point cloud, n d >> n s ;

[0031] Step 52: Use the Delaunay triangulation algorithm to convert the dense point cloud into a triangular mesh. The calculation formula for the triangulation process of the Delaunay triangulation algorithm is:

[0032] T = Delaunay(P d )

[0033] where T = {Δ k | k = 1, 2, 3, ..., m} is the set of triangular meshes generated by Delaunay triangulation. Each triangle Δ k is composed of three points;

[0034] Step 53: Optimize the triangular mesh to improve the visual effect and geometric accuracy of the model. The expression for the mesh optimization process is:

[0035] T * = Optimize(T)

[0036] The initial set of triangular meshes is T = {△ k}, and the optimized set of triangular meshes is

[0037] Step 54: Map the texture information of the image dataset onto the optimized triangular mesh to endow the model with a sense of reality and improve the accuracy of subsequent segmentation of the landslide area of the model using the segmentation algorithm. Obtain the 3D model;

[0038] The texture mapping expression is:

[0039]

[0040] where is the optimized set of triangular meshes, is the 3D coordinate of the j-th vertex, and Texture(T * ) is the set of textured triangular meshes.

[0041] For the self-built aerial image dataset of the woodland landslide and collapse area, the SFM method is used to reconstruct the sparse three-dimensional point cloud, and the refined three-dimensional model is constructed by combining with the computer vision openMVS method.

[0042] Preferably, in step 6, the area recognition and calculation module segments the woodland landslide and collapse area in the three-dimensional model through the SAM algorithm for universal segmentation. The specific steps are as follows:

[0043] Step 61: The area recognition and calculation module generates two-dimensional images I from multiple perspectives by transforming the three-dimensional model through the projection matrix. Each two-dimensional image I corresponds to a part of the three-dimensional model. The three-dimensional coordinates are converted into two-dimensional coordinates through the projection matrix to obtain the two-dimensional image I. The expression of the projection transformation process is: two , each two-dimensional image I two corresponds to a part of the three-dimensional model; the three-dimensional coordinates are converted into two-dimensional coordinates through the projection matrix to obtain the two-dimensional image I two , and the expression of the projection transformation process is:

[0044] Projected coordinates = P × X

[0045] where P is the projection matrix and X is the point in the three-dimensional model;

[0046] Step 62: Use the SAM algorithm for universal segmentation to perform semantic segmentation on the two-dimensional image I, and segment and output the landslide area data S under each perspective. The function expression of the SAM algorithm for universal segmentation is: two S = SAM(I

[0047] S = SAM(I two )

[0048] where SAM() is the universal segmentation algorithm;

[0049] Step 63: Reverse map the landslide area data S to the three-dimensional space to obtain the three-dimensional coordinates X. The reverse mapping expression is: s , and the reverse mapping expression is:

[0050]

[0051] where (x s , y s ) is the coordinate of the landslide area pixel in the two-dimensional image I two , and P -1 is the reverse projection matrix;

[0052] Step 64: Combine the three-dimensional landslide area markers X from each perspective to generate the final three-dimensional landslide area marker X. The calculation expression for combination is: s,i Combine the three-dimensional landslide area markers X from each perspective to generate the final three-dimensional landslide area marker X combined , and the calculation expression for combination is:

[0053]

[0054] Through the above design, the three-dimensional model of the woodland landslide and collapse area is mapped to two-dimensional planes from multiple perspectives. The woodland landslide and collapse area is segmented by the SAM algorithm, and the segmented two-dimensional results are reflected back to the three-dimensional space to generate the final three-dimensional landslide area.

[0055] Preferably, in step 7, the Delaunay triangulation algorithm is used to divide the landslide area into L triangular elements, where L≥2. The vertex coordinates and corresponding side lengths of each triangular element are recorded, and the area A of each triangular element is calculated using Heron's formula. i , and the expression is as follows:

[0056]

[0057] where, a, b, and c are the side lengths of the triangle;

[0058] The areas of all triangular elements are accumulated to obtain the total area A of the woodland landslide area. total :

[0059]

[0060] where N is the total number of triangles in the landslide area.

[0061] Through the above design, based on the determined three-dimensional area of the woodland landslide and collapse area, the corresponding triangular elements of this area are found by the Delaunay method. Combining Heron's formula to calculate the areas of the triangular elements within the area, and accumulating the sum of the areas of all triangles to obtain the area of the woodland landslide and collapse.

[0062] Advantages of the present invention: The present invention obtains high-resolution images from multiple perspectives by using drones, generates a three-dimensional model using three-dimensional reconstruction technology, segments the landslide and collapse area based on a deep learning segmentation algorithm, and finally realizes the accurate area detection of the woodland landslide and collapse area through the calculation of the triangular mesh area. It has high-resolution and high-precision feature points, greatly improving the area monitoring efficiency of the woodland landslide and collapse area in natural disasters, and can provide important technical support for the early warning and prevention of woodland landslide and collapse disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 is a schematic flow chart of the present invention;

[0064] Figure 2 is a schematic diagram of dividing the landslide area into L triangular elements in the embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0065] The present invention will be further described in detail below with reference to the drawings and specific examples. The following examples or drawings are used to illustrate the present invention, but do not limit the scope of the present invention.

[0066] As shown in Figure 1 the following: A method for detecting the landslide collapse area of forest land based on multi-view UAV remote sensing, comprising the following steps:

[0067] Step 1: Construct a system for detecting the landslide collapse area of forest land based on multi-view UAV remote sensing. The system for detecting the landslide collapse area of forest land includes a UAV aircraft, which is equipped with a GPS, a radar, and a high-definition camera. A route planning module, a 3D model construction module, and a region recognition and calculation module are also set in the UAV aircraft;

[0068] Step 2: The route planning module plans the aerial photography route for the image acquisition task of the forest land target area;

[0069] Step 3: The UAV aircraft takes images of the forest land target area according to the aerial photography route planned by the route planning module to obtain an image data set;

[0070] Step 4: The 3D model construction module uses the SFM method to perform sparse point cloud reconstruction on the image data set to obtain the sparse point cloud of the forest land target area;

[0071] Step 5: The 3D model construction module uses the MVS method to perform high-precision 3D reconstruction on the sparse point cloud to obtain a refined 3D model of the target area;

[0072] Step 6: The region recognition and calculation module segments the landslide collapse area of the forest land in the 3D model through the SAM algorithm for universal segmentation to determine the landslide area data of the forest land;

[0073] Step 7: The region recognition and calculation module, based on the landslide area data of the forest land, obtains all the triangular elements used in the triangulation modeling for each landslide area of the forest land, then uses Heron's formula to calculate the triangular area of each landslide area of the forest land, and finally accumulates the triangular areas of all the landslide areas of the forest land to obtain the total area data of the landslide area of the forest land.

[0074] In the step 2, the route planning module uses the five-direction flight method to plan the aerial photography route. In this embodiment, the five-direction flight method is specifically as follows: the route overlap rate is 85%, the side overlap rate is 80%, the orthophoto route shooting angle is 90 degrees, the oblique photography shooting angle is 45 degrees, and the flight mode uses terrain-following flight with a height difference of 15 m. To ensure the fineness of the subsequent 3D reconstruction model.

[0075] In the step 4, the 3D model construction module uses the SFM method to perform sparse point cloud reconstruction on the image data set, and the specific steps are as follows:

[0076] Step 41: The 3D model construction module extracts the feature points of each image data using the SIFT algorithm, finds the key points of the image, and generates the descriptors related to each feature point; then uses the Fast Library for Approximate Nearest Neighbors (FLANN) to match the extracted feature points to obtain the matching point pairs. The approximate nearest neighbor algorithm can accelerate the matching between feature points, and calculate the Euclidean distance d between the descriptor vectors corresponding to each matching point pair.

[0077] The calculation formula for the Euclidean distance of the matching point pairs is as follows:

[0078]

[0079] where p i and q i are the descriptor vectors of the two feature points in the matching point pair;

[0080] Step 42: Based on the matching point pairs, the sparse point cloud is reconstructed through the Structure from Motion (SFM) algorithm. The calculation formula is as follows:

[0081]

[0082] where P1 and P2 are the projection matrices of two cameras respectively, x1 and x2 are the pixel coordinates of the matching feature points on two images, and X p is the 3D coordinate point in the sparse point cloud.

[0083] In step 5, the 3D model construction module uses the Multi-View Stereo (MVS) method to perform high-precision 3D reconstruction on the sparse point cloud. The specific steps are as follows:

[0084] Step 51: Based on the sparse point cloud, the 3D model construction module uses the MVS method to generate a high-density point cloud, that is, by performing geometric reasoning on the multi-view image dataset, the sparse point cloud is expanded into a dense point cloud;

[0085] The expression for point cloud reconstruction using the MVS method is:

[0086] P d = MVS(P s , I)

[0087] where I is the multi-view image dataset, P s = {p i | p i ∈ R 3 , i = 1, 2,..., n s} is the input sparse point cloud, n s is the number of points in the sparse point cloud, and P d = {q j | q j ∈ R3 , j = 1, 2, ..., n d} is the output dense point cloud, n d is the number of points in the dense point cloud, n d >> n s ;

[0088] Step 52: Use the Delaunay triangulation algorithm to convert the dense point cloud into a triangular mesh. The calculation formula for the triangulation process of the Delaunay triangulation algorithm is:

[0089] T = Delaunay(P d )

[0090] where T = {Δ k | k = 1, 2, 3, ..., m} is the set of triangular meshes generated by Delaunay triangulation. Each triangle Δ k is composed of three points;

[0091] Step 53: Optimize the triangular mesh to improve the visual effect and geometric accuracy of the model. The expression for the mesh optimization process is:

[0092] T * = Optimize(T)

[0093] The initial set of triangular meshes is T = {△ k}}, and the optimized set of triangular meshes is

[0094] Step 54: Map the texture information of the image dataset onto the optimized triangular mesh to endow the model with a sense of reality and improve the accuracy of subsequent segmentation of the landslide area of the model using the segmentation algorithm. Obtain the 3D model;

[0095] The texture mapping expression is:

[0096]

[0097] where is the optimized set of triangular meshes, is the 3D coordinate of the j-th vertex, and Texture(T * ) is the set of textured triangular meshes.

[0098] In step 6, the region recognition calculation module segments the forest land landslide and collapse area in the 3D model through the SAM algorithm for universal segmentation. The specific steps are as follows:

[0099] Step 61: The region recognition and calculation module generates two-dimensional images I of multiple perspectives by transforming the three-dimensional model through a projection matrix. Each two-dimensional image I corresponds to a part of the three-dimensional model. The three-dimensional coordinates are converted into two-dimensional coordinates through the projection matrix to obtain the two-dimensional image I. The expression of the projection transformation process is: Projection coordinates = P × X, where P is the projection matrix and X is a point in the three-dimensional model. two , each two-dimensional image I two corresponds to a part of the three-dimensional model; the three-dimensional coordinates are converted into two-dimensional coordinates through the projection matrix to obtain the two-dimensional image I two , and the expression of the projection transformation process is:

[0100] Projection coordinates = P × X

[0101] where P is the projection matrix and X is a point in the three-dimensional model;

[0102] Step 62: Use the Segment Anything Model (SAM) algorithm to perform semantic segmentation on the two-dimensional image I, and segment and output the landslide area data S under each perspective. The function expression of the SAM algorithm is: S = SAM(I), where SAM() is the Segment Anything algorithm. two Perform semantic segmentation on it, and segment and output the landslide area data S under each perspective; the function expression of the SAM algorithm is:

[0103] S = SAM(I two )

[0104] where SAM() is the Segment Anything algorithm;

[0105] Step 63: Reverse map the landslide area data S to the three-dimensional space to obtain the three-dimensional coordinates X. The reverse mapping expression is: where (x, y) are the coordinates of the pixels in the landslide area of the two-dimensional image I, and P is the inverse projection matrix. s The reverse mapping expression is:

[0106]

[0107] where, (x s , y s ) are the coordinates of the pixels in the landslide area of the two-dimensional image I two , and P -1 is the inverse projection matrix;

[0108] Step 64: Combine the three-dimensional landslide area markers X of each perspective to generate the final three-dimensional landslide area marker X. The calculation expression for combination is: s,i Combine the three-dimensional landslide area markers X of each perspective to generate the final three-dimensional landslide area marker X combined , and the calculation expression for combination is:

[0109]

[0110] In step 7, use the Delaunay triangulation algorithm to divide the landslide area into L triangular cells, as shown in, record the vertex coordinates and corresponding side lengths of each triangular cell, and use Heron's formula to calculate the area A of each triangular cell. The expression is as follows: Figure 2 As shown, record the vertex coordinates and corresponding side lengths of each triangular cell, and use Heron's formula to calculate the area A of each triangular cell i , and the expression is as follows:

[0111]

[0112] where, a, b, and c are the side lengths of a triangle;

[0113] Accumulate the areas of all triangular elements to obtain the total area A of the forest landslide area total :

[0114]

[0115] where N is the total number of triangles in the landslide area.

[0116] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for detecting forest landslide collapse area based on multi-view UAV remote sensing, characterized in that: The following steps are involved: Step 1: construct a forest landslide collapse area detection system based on multi-view UAV remote sensing, wherein the forest landslide collapse area detection system includes a UAV aircraft, which is equipped with GPS, radar and high-definition camera, and is also provided with a route planning module, a three-dimensional model construction module and a region identification calculation module; Step 2: The route planning module performs aerial photography route planning for the image acquisition task in the forest target area; Step 3: The UAV aircraft takes images of the forest target area according to the aerial photography route planned by the route planning module to obtain an image data set; Step 4: The three-dimensional model construction module uses the SFM method to reconstruct the sparse point cloud of the image data set to obtain a sparse point cloud of the forest target area; Step 5: The 3D model building module uses the MVS method to perform high-precision 3D reconstruction on the sparse point cloud to obtain a refined 3D model of the target area; Step 6: The area recognition calculation module segments the forest landslide collapse area in the three-dimensional model through the SAM algorithm to determine the forest landslide area data; Step 7: The area identification and calculation module obtains all the triangular elements used in the triangulated modeling of each forest landslide area based on the forest landslide area data, and then uses Heron's formula to calculate the triangular area of ​​each forest landslide area. Finally, the triangular areas of all forest landslide areas are accumulated to obtain the total area data of the forest landslide area.

2. The method for detecting forest landslide collapse area based on multi-view UAV remote sensing according to claim 1 is characterized in that: In step 2, the route planning module adopts a five-way flight method for aerial photography route planning. Specifically, the route overlap rate is 85%, the lateral overlap rate is 80%, the orthographic route shooting angle is 90 degrees, the oblique photography shooting angle is 45 degrees, and the flight mode adopts a terrain-simulating flight with a height difference of 15m.

3. The method for detecting forest landslide collapse area based on multi-view UAV remote sensing according to claim 1 is characterized in that: In step 4, the three-dimensional model construction module uses the SFM method to reconstruct the sparse point cloud of the image data set. The specific steps are as follows: Step 41: The three-dimensional model building module uses the SIFT algorithm to extract feature points of each image data and generates descriptors related to each feature point; then uses the fast nearest neighbor search FLANN to match the extracted feature points to obtain matching point pairs, and calculates the Euclidean distance d between the descriptor vectors corresponding to each matching point pair; The Euclidean distance calculation formula for matching point pairs is as follows: Among them, p i and q i is the descriptor vector of the two feature points in the matching point pair; Step 42: Reconstruct the sparse point cloud based on the matching point pairs using the SFM algorithm. The calculation formula is as follows: Among them, P1 and P2 are the projection matrices of the two cameras, x1 and x2 are the pixel coordinates of the matching feature points on the two images, and X p is the 3D coordinate point in the sparse point cloud.

4. The method for detecting forest landslide collapse area based on multi-view UAV remote sensing according to claim 1, characterized in that: In step 5, the three-dimensional model construction module uses the MVS method to perform high-precision three-dimensional reconstruction on the sparse point cloud. The specific steps are as follows: Step 51: The three-dimensional model construction module generates a high-density point cloud based on the sparse point cloud using the MVS method, that is, the sparse point cloud is expanded into a dense point cloud by performing geometric reasoning on the multi-view image data set; The expression for point cloud reconstruction using the MVS method is: P d =MVS(P s ,I) Among them, I is a multi-view image dataset, P s ={p i |p i ∈R 3 ,i=1,2,...,n s } is the input sparse point cloud, n s is the number of points in the sparse point cloud, P d = {q j |q j ∈R 3 ,j=1,2,...,n d } is the output dense point cloud, n d is the number of points in the dense point cloud, n d >>n s ; Step 52: The dense point cloud is converted into a triangular mesh using the Delaunay triangulation algorithm. The triangulation process calculation formula of the Delaunay triangulation algorithm is: T=Delaunay(P d ) Where T = {Δ k |k=1,2,3,...,m} is a set of triangular meshes generated by Delaunay triangulation algorithm, and each triangle Δ k It is composed of three points; Step 53: Optimize the triangular mesh. The expression of the mesh optimization process is: T * =Optimize(T) The initial triangular mesh set is T = {△ k }, the optimized triangular mesh set is Step 54: Mapping the texture information of the image data set onto the optimized triangular mesh to obtain the three-dimensional model; The texture mapping expression is: in, is the optimized triangle mesh set, is the three-dimensional coordinate of the j-th vertex, Texture(T * ) is a set of textured triangle meshes.

5. The method for detecting forest landslide collapse area based on multi-view UAV remote sensing according to claim 1, characterized in that: In step 6, the region recognition calculation module segments the forest landslide collapse area in the three-dimensional model through the SAM algorithm, and the specific steps are as follows: Step 61: The region recognition calculation module transforms the three-dimensional model into two-dimensional images of multiple viewing angles using a projection matrix. two , each two-dimensional image I two Corresponding to a part of the three-dimensional model; the expression of the projection transformation process is: Projection coordinates = P × X Where P is the projection matrix and X is the point in the three-dimensional model; Step 62: Use the SAM algorithm to segment the two-dimensional image I two Semantic segmentation is performed to segment and output the landslide area data S under each perspective; the function expression of the SAM algorithm for all-things segmentation is: S=SAM(I two ) Among them, SAM() is the all things segmentation algorithm; Step 63: Reversely map the landslide area data S to three-dimensional space to obtain three-dimensional coordinates X s , the reverse mapping expression is: Among them, (x s ,y s ) is a two-dimensional image I two The coordinates of the pixels in the landslide area, P -1 is the inverse projection matrix; Step 64: Mark the 3D landslide area in each view with an X s,i Combine and generate the final three-dimensional landslide area mark X combined , the calculation expression of the combination is:

6. The method for detecting forest landslide collapse area based on multi-view UAV remote sensing according to claim 1, characterized in that: In step 7, the landslide area is divided into L triangular units using the Delaunay triangulation algorithm, the vertex coordinates and corresponding side lengths of each triangular unit are recorded, and the area A of each triangular unit is calculated using the Heron formula. i , the expression is as follows: in, a, b, c are the lengths of the sides of the triangle; The areas of all triangular units are accumulated to obtain the total area A of the forest landslide area. total : Where N is the total number of triangles in the landslide area.

Images