Forest land landslide collapse area detection method based on unmanned aerial vehicle visual remote sensing

Abstract

A forest land landslide collapse area detection method based on unmanned aerial vehicle visual remote sensing is characterized by comprising the following steps: step 1, constructing a forest land landslide collapse area detection system based on unmanned aerial vehicle visual remote sensing; 2, the unmanned aerial vehicle shoots remote sensing image data; 3, an image preprocessing module performs preprocessing operation on the historical satellite image data and the remote sensing image data to obtain a remote sensing image and a historical image; step 4, matching the feature points of the remote sensing image and the historical image by an image alignment module, calculating a transformation matrix, and completing alignment of the images before and after the landslide through the transformation matrix; step 5, identifying a landslide area by a landslide area identification module; step 6, a landslide area identification module extracts a landslide area contour; and 7, calculating the average gradient of the landslide area by a landslide area calculation module, and calculating the actual area of the slope of the landslide in combination with the plane projection area and the average gradient. The method has the advantage that the forest land landslide collapse area can be rapidly and accurately detected.

Description

Technical Field

[0001] The present invention relates to the technical field of forest landslide detection, and in particular to a method for detecting the collapse area of ​​forest landslide based on unmanned aerial vehicle visual remote sensing. Background Art

[0002] Landslides in forests are a common geological disaster that poses a serious threat to forest ecosystems and human society. Landslides in forests not only cause large-scale destruction of vegetation cover, but may also trigger secondary disasters such as mudslides, posing a major risk to the surrounding environment and the safety of residents. Timely and accurate detection and monitoring of the scope and area of ​​landslides in forests is of vital importance for assessing the impact of disasters, formulating prevention and control measures, and guiding ecological restoration work.

[0003] Accurate measurement of landslide area is a key link in disaster assessment. It can not only intuitively reflect the scale and intensity of landslides, but also provide an important basis for subsequent ecological restoration and geological governance. 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 and provide scientific support for the establishment and improvement of early warning systems.

[0004] Traditional methods for detecting forest landslide areas mainly include field measurement and satellite remote sensing. The field measurement method requires professionals to enter the landslide area for field surveys and measurements. Although this method can obtain relatively accurate data, it has the following disadvantages: first, the operation process is time-consuming and laborious, especially in forest areas with complex terrain and inconvenient transportation; second, field measurement may face safety risks, especially when the landslide has just occurred or is still ongoing; finally, the efficiency of field measurement is low, and it is difficult to meet the needs of rapid assessment of large-scale or multiple landslide disasters.

[0005] Although satellite remote sensing technology has overcome some limitations of field measurements, it also has obvious shortcomings. High-resolution satellite images are expensive to obtain and are limited by orbital cycles and weather conditions, making it difficult to achieve timely monitoring of landslide areas. Although low-resolution satellite images are easier to obtain, their spatial resolution is often insufficient to accurately identify and measure small landslides. In addition, the application of satellite remote sensing is limited in densely forested areas, and it is difficult to penetrate dense canopy to obtain surface information.

[0006] Therefore, the existing technologies have the following major defects in detecting the area of ​​landslide collapse in forests: the field measurement method is time-consuming and labor-intensive, has safety risks and is inefficient; the satellite remote sensing method is limited by high cost, low timeliness and application limitations in complex terrain and dense forest areas. These problems seriously restrict the rapid assessment and effective management of forest landslide disasters. Summary of the invention

[0007] The present invention provides a method for detecting the area of ​​forest landslide and collapse based on unmanned aerial vehicle visual remote sensing, which can quickly, accurately, safely, economically and efficiently realize the precise detection of the area of ​​forest landslide and collapse.

[0008] To achieve the above-mentioned purpose, the present invention provides a method for detecting forest landslide collapse area based on UAV visual remote sensing, which comprises the following steps:

[0009] Step 1: System construction: Construct a forest landslide collapse area detection system based on UAV visual remote sensing, the forest landslide collapse area detection system includes a UAV, the UAV is equipped with a multispectral camera, and the multispectral camera is sequentially connected to an image preprocessing module, an image alignment module, a landslide area recognition module and a landslide area calculation module;

[0010] Step 2: Data acquisition: The UAV is equipped with a multispectral camera to take remote sensing images of the forest landslide area, obtain remote sensing image data, and pass it to the image preprocessing module;

[0011] Step 3: Image preprocessing: The image preprocessing module obtains the historical satellite image data of the forest landslide area, and performs preprocessing operations on the historical satellite image data and remote sensing image data to obtain standard remote sensing images and historical images, and pass them to the image alignment module;

[0012] Step 4: Image alignment: The image alignment module extracts feature points from the remote sensing image and the historical image, matches the feature points of the two, and then calculates a transformation matrix based on the matched feature point pairs, and maps the remote sensing image and the historical image to the same spatial reference system through the transformation matrix, thereby completing the alignment of the images before and after the landslide;

[0013] Step 5: Landslide area identification: the landslide area identification module identifies the landslide area by calculating the ratio vegetation index RVI of the remote sensing image and the historical image;

[0014] Step 6: Image post-processing: The landslide area recognition module uses the corrosion and expansion operations of OpenCV to remove small noises and connect the scattered landslide areas to extract the contour of the landslide area;

[0015] Step 7: Calculation of landslide area: The landslide area calculation module uses the GDAL tool to perform coordinate conversion, generate a digital elevation model DEM, calculate the average slope of the landslide area, and then combine the plane projection area and the average slope to calculate the actual area of ​​the landslide slope to obtain the forest landslide collapse area.

[0016] Through the above design, firstly, a multispectral camera equipped with an unmanned aerial vehicle is used to obtain remote sensing images of the landslide area, and historical satellite images before the landslide are collected. Then, data preprocessing is performed, including radiation calibration and image correction operations. Next, the SIFT algorithm is used to align the images before and after the landslide to ensure the consistency of the spatial position of the images. Subsequently, the bare soil area after the landslide is identified by the ratio vegetation index RVI, and the landslide area is determined according to the change of the RVI value. Finally, the landslide area is eroded and expanded to extract the final landslide area contour, so as to realize the accurate identification and area calculation of the landslide area. This method can efficiently, accurately, safely and cost-effectively identify and calculate the area of ​​forest landslide collapse, providing a scientific basis for disaster assessment and the formulation of prevention and control measures.

[0017] Preferably, in step 2, a multispectral camera is mounted on an unmanned aerial vehicle to take multiple overlapping remote sensing images of the forest landslide area, with the heading overlap rate and the lateral overlap rate being not less than 80%, for subsequent reconstruction of the three-dimensional point cloud of the landslide area to obtain an elevation model.

[0018] At the same time, historical satellite remote sensing images before the landslide occurred are collected to provide basic data for subsequent reconstruction and comparative analysis of the landslide area.

[0019] Preferably, in step 3, the image preprocessing module performs preprocessing operations on the remote sensing image data and the historical satellite image data, and the specific steps are as follows:

[0020] Step 31: Radiation calibration: The image preprocessing module converts the remote sensing image data and historical satellite image data stored in the form of digital values ​​DN into the form of radiation brightness values ​​through a conversion formula. The conversion formula is as follows:

[0021] L=(DN-DN offset )×Gain

[0022] Where L is the radiant brightness value, DN is the raw digital value, and DN offset is the offset value of the sensor, Gain is the gain coefficient;

[0023] The original image data captured by the remote sensor is stored in the form of digital values, namely DN values, which are converted into radiant brightness values, namely the radiant energy per unit area, unit solid angle and unit wavelength, through a conversion formula, so as to facilitate the subsequent image correction step to correct the atmospheric apparent reflectance to the surface reflectance.

[0024] Step 32: Image correction: The image preprocessing module performs processing operations including but not limited to geometric correction, image fusion, image cropping, and shadow removal on the remote sensing image data and historical satellite image data in the form of radiation brightness values, and then uses the FLAASH atmospheric correction model to correct the radiation reflectivity to the surface reflectivity.

[0025] In the process of remote sensing imaging, factors such as sensor instability and water vapor and aerosol scattering effects may cause geometric distortion and radiation distortion of images. These factors have a serious impact on the quality of remote sensing images. In order to solve the geometric distortion problem caused by sensor instability, geometric correction (including coarse correction, geometric fine correction and orthorectification), image fusion, image cropping and shadow removal are used to improve the quality of remote sensing images; in order to solve the radiation distortion caused by water vapor and aerosol scattering effects, the FLAASH atmospheric correction model is used to remove the atmospheric influence, calculate the atmospheric apparent reflectance, and then correct the atmospheric apparent reflectance to the surface reflectance to reflect the true reflectance of the surface.

[0026] Preferably, the geometric correction includes coarse correction, fine geometric correction and orthorectification.

[0027] Preferably, in step 4, the image alignment module uses the SIFT scale-invariant feature transformation method to detect feature points in the remote sensing image and the historical image, and then matches the feature points of the remote sensing image and the historical image through a nearest neighbor search algorithm to obtain a set of feature point pairs.

[0028] The purpose of remote sensing image alignment is to accurately align the satellite image before the landslide with the drone remote sensing image after the landslide, so as to ensure that the same geographical location has a consistent spatial position in the images at different times. The SIFT scale-invariant feature transformation method is used to detect the feature points in the historical image and the remote sensing image, and the feature descriptor of each feature point is generated. The nearest neighbor search algorithm is used to find the corresponding feature point pairs in the historical image before the landslide and the remote sensing image after the landslide. The transformation matrix is ​​calculated by matching the feature point pairs, which maps the images before and after the landslide to the same spatial reference system.

[0029] Preferably, in step 5, the landslide area identification module uses the cross-platform computer vision and machine learning software library OpenCV to read remote sensing images and extract data from the red light band and the near infrared band; the calculation expression of the ratio vegetation index RVI of the remote sensing image is as follows:

[0030]

[0031] Among them, I RED is the red light band data value of the corresponding pixel in the remote sensing image, I NIR It is the near-infrared band data value of the corresponding pixel in the remote sensing image.

[0032] The ratio vegetation index (RVI) is calculated by the ratio of the red band and the near-infrared band. It can be used to distinguish between vegetation and soil, and then the ratio vegetation index (RVI) can be used to identify the mud and rock areas produced after the landslide.

[0033] An RVI threshold T is set to distinguish soil and non-soil areas. RVI values ​​in the range of 0-2 are identified as bare soil and rock areas, and RVI values ​​in the range of 2-8 are identified as green vegetation.

[0034] The occurrence of forest landslides will cause the forest vegetation area to be covered by bare soil. The area reflected by the RVI value as vegetation before the landslide will have a lower RVI value after the landslide. For the case where the forest vegetation area is covered by landslide debris, the RVI value will be reflected in the RVI value as the RVI value before the landslide is between 2 and 8, and the RVI value after the landslide is between 0 and 2. Therefore, the area where the forest landslide occurs is determined based on the change in the RVI value.

[0035] Preferably, in step 6, the landslide area identification module uses the corrosion function in OpenCV to perform corrosion and expansion processing on the identified landslide area, and the specific steps are as follows:

[0036] Step 61: Select a 3X3 erosion convolution kernel and use the cv2.erode() function in OpenCV to perform an erosion operation on the landslide area image to obtain an eroded image;

[0037] The erosion operation can remove small noise points in the landslide area and eliminate small patches that are not connected to the main landslide area. The erosion operation slides the convolution kernel in the image and replaces the pixel values ​​in the area with the local minimum, thereby shrinking the area.

[0038] Step 62: Select a 3X3 dilated convolution kernel and use the cv2.dilate() function in OpenCV to dilate the eroded image to obtain a dilated image.

[0039] Dilation can expand and connect scattered landslide areas to make them more coherent, while restoring major landslide areas that may have been overly reduced during the erosion process. The dilation operation slides the convolution kernel across the image, replacing the pixel values ​​in the region with the local maximum, thereby expanding the region.

[0040] Step 63: Use the contour extraction function cv2.findContours() in OpenCV to extract the contour of the expanded image and obtain the contour of the landslide area, which is convenient for subsequent calculation of the area of ​​the landslide area or other analysis.

[0041] Preferably, in step 7, the landslide area calculation module calculates the forest landslide collapse area, and the specific steps are as follows:

[0042] Step 71: The landslide area calculation module uses the `gdaltransform` tool in the open source raster spatial data conversion library GDAL to convert the pixel coordinates (rows, columns) in the remote sensing image into geographic coordinates (longitude, latitude); uses the `ogr2ogr` tool in the open source raster spatial data conversion library GDAL to save the converted landslide boundary geographic coordinates into a vector format, such as Shapefile or GeoJSON, to facilitate further analysis and visualization in GIS software.

[0043] Step 72: Use the 3D reconstruction open source software library OpenMVG to perform camera pose estimation and sparse point cloud generation. After the sparse point cloud is generated, further optimize the reconstruction model and generate a dense point cloud, and then use the GDAL tool to convert the dense point cloud into a digital elevation model DEM; each point of the dense point cloud contains X, Y and Z coordinates, and use the `gdal_grid` command in GDAL to interpolate each point of the dense point cloud to generate regular DEM grid data;

[0044] Step 73: According to the DEM grid data, the slope is calculated using the gdaldem command in the GDAL tool; a mask is created to extract the slope data of the landslide area in the remote sensing image, and after all the slope values ​​in the landslide area are extracted, the average value of all the slope values ​​is calculated. The slope average value calculation formula is as follows:

[0045]

[0046] Where N is the total number of pixels in the landslide area, θ i is the slope value of the i-th pixel, in degrees;

[0047] Step 74: According to the landslide area contour, obtain the coordinates (x i ,y i ), and then calculate the projection area of ​​the landslide area on the horizontal plane. The calculation formula is as follows:

[0048]

[0049] Among them, (x i ,y i ) is the plane coordinate of the i-th vertex of the polygon, and n is the number of vertices of the polygon;

[0050] The average slope is then used to calculate the actual area of ​​the landslide slope, that is, the area of ​​forest landslide collapse, and the calculation formula is as follows:

[0051]

[0052] Among them, A 平面is the calculated plane projection area, and θ is the average slope of the landslide area.

[0053] The beneficial effects of the present invention are as follows: the area of ​​forest landslide collapse can be identified and calculated efficiently, accurately, safely and economically, and the precise monitoring of the range of forest landslide collapse can be achieved, aiming to provide a scientific basis for landslide disaster assessment, formulation of prevention and control measures and ecological restoration planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0055] The present invention is further described in detail below in conjunction with the accompanying drawings and specific examples. The following examples or drawings are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0056] like Figure 1 As shown: A method for detecting forest landslide collapse area based on UAV visual remote sensing, comprising the following steps:

[0057] Step 1: System construction: Construct a forest landslide collapse area detection system based on UAV visual remote sensing, the forest landslide collapse area detection system includes a UAV, the UAV is equipped with a multispectral camera, and the multispectral camera is sequentially connected to an image preprocessing module, an image alignment module, a landslide area recognition module and a landslide area calculation module;

[0058] Step 2: Data acquisition: The UAV is equipped with a multispectral camera to take remote sensing images of the forest landslide area, obtain remote sensing image data, and pass it to the image preprocessing module;

[0059] Step 3: Image preprocessing: The image preprocessing module obtains the historical satellite image data of the forest landslide area, and performs preprocessing operations on the historical satellite image data and remote sensing image data to obtain standard remote sensing images and historical images, and pass them to the image alignment module;

[0060] Step 4: Image alignment: The image alignment module extracts feature points from the remote sensing image and the historical image, matches the feature points of the two, and then calculates a transformation matrix based on the matched feature point pairs, and maps the remote sensing image and the historical image to the same spatial reference system through the transformation matrix, thereby completing the alignment of the images before and after the landslide;

[0061] Step 5: Landslide area identification: the landslide area identification module identifies the landslide area by calculating the ratio vegetation index RVI of the remote sensing image and the historical image;

[0062] Step 6: Image post-processing: The landslide area recognition module uses the corrosion and expansion operations of OpenCV to remove small noises and connect the scattered landslide areas to extract the contour of the landslide area;

[0063] Step 7: Calculation of landslide area: The landslide area calculation module uses the GDAL tool to perform coordinate conversion, generate a digital elevation model DEM, calculate the average slope of the landslide area, and then combine the plane projection area and the average slope to calculate the actual area of ​​the landslide slope to obtain the forest landslide collapse area.

[0064] In step 2, a multispectral camera is used to take multiple overlapping remote sensing images of the forest landslide area using an unmanned aerial vehicle, with the heading overlap rate and the lateral overlap rate not less than 80%. This is used to subsequently reconstruct the three-dimensional point cloud of the landslide area and obtain an elevation model; in addition, historical satellite images before the forest landslide occurred need to be obtained.

[0065] In step 3, the image preprocessing module performs preprocessing operations on the remote sensing image data and the historical satellite image data. The specific steps are as follows:

[0066] Step 31: Radiation calibration: The image preprocessing module converts the remote sensing image data and historical satellite image data stored in the form of digital values ​​DN into the form of radiation brightness values ​​through a conversion formula. The conversion formula is as follows:

[0067] L=(DN-DN offset )×Gain

[0068] Where L is the radiant brightness value, DN is the raw digital value, and DN offset is the offset value of the sensor, Gain is the gain coefficient;

[0069] Step 32: Image correction: The image preprocessing module performs geometric correction, image fusion, image cropping, shadow removal and other processing operations on the remote sensing image data and historical satellite image data in the form of radiation brightness values, and then uses the FLAASH atmospheric correction model to correct the radiation reflectivity to the surface reflectivity.

[0070] The geometric correction includes coarse correction, fine geometric correction and orthorectification.

[0071] In step 4, the image alignment module uses the SIFT scale-invariant feature transformation method to detect feature points in the remote sensing image and the historical image, and then matches the feature points of the remote sensing image and the historical image through a nearest neighbor search algorithm to obtain a set of feature point pairs.

[0072] In step 5, the landslide area identification module uses the cross-platform computer vision and machine learning software library OpenCV to read the remote sensing image and extract the data of the red light band and the near infrared band; the calculation expression of the ratio vegetation index RVI of the remote sensing image is as follows:

[0073]

[0074] Among them, I RED is the red light band data value of the corresponding pixel in the remote sensing image, I NIR It is the near-infrared band data value of the corresponding pixel in the remote sensing image.

[0075] In this embodiment, the RVI threshold T=2 is set to distinguish soil and non-soil areas. RVI values ​​in the range of 0-2 are identified as bare soil and rock areas, and RVI values ​​in the range of 2-8 are identified as green vegetation.

[0076] The occurrence of forest landslides will cause the forest vegetation area to be covered by bare soil. The area reflected by the RVI value as vegetation before the landslide will have a lower RVI value after the landslide. For the case where the forest vegetation area is covered by landslide debris, the RVI value will be reflected in the RVI value as the RVI value before the landslide is between 2 and 8, and the RVI value after the landslide is between 0 and 2. Therefore, the area where the forest landslide occurs is determined based on the change in the RVI value.

[0077] In step 6, the landslide area identification module uses the erosion function in OpenCV to perform erosion and expansion processing on the identified landslide area. The specific steps are as follows:

[0078] Step 61: Select a 3X3 erosion convolution kernel and use the cv2.erode() function in OpenCV to perform an erosion operation on the landslide area image to obtain an eroded image;

[0079] Step 62: Select a 3X3 dilated convolution kernel and use the cv2.dilate() function in OpenCV to dilate the eroded image to obtain a dilated image.

[0080] Step 63: Use the contour extraction function cv2.findContours() in OpenCV to extract the contour of the expanded image and obtain the contour of the landslide area.

[0081] In step 7, the landslide area calculation module calculates the forest landslide collapse area, and the specific steps are as follows:

[0082] Step 71: The landslide area calculation module uses the `gdaltransform` tool in the open source raster spatial data conversion library GDAL to convert the pixel coordinates (rows, columns) in the remote sensing image into geographic coordinates (longitude, latitude); uses the `ogr2ogr` tool in the open source raster spatial data conversion library GDAL to save the converted landslide boundary geographic coordinates into a vector format; such as Shapefile or GeoJSON, to facilitate further analysis and visualization in GIS software.

[0083] Step 72: Use the 3D reconstruction open source software library OpenMVG to perform camera pose estimation and sparse point cloud generation. After the sparse point cloud is generated, further optimize the reconstruction model and generate a dense point cloud, and then use the GDAL tool to convert the dense point cloud into a digital elevation model DEM; each point of the dense point cloud contains X, Y and Z coordinates, and use the `gdal_grid` command in GDAL to interpolate each point of the dense point cloud to generate regular DEM grid data;

[0084] Step 73: According to the DEM grid data, the slope is calculated using the gdaldem command in the GDAL tool; a mask is created to extract the slope data of the landslide area in the remote sensing image, and after all the slope values ​​in the landslide area are extracted, the average value of all the slope values ​​is calculated. The slope average value calculation formula is as follows:

[0085]

[0086] Where N is the total number of pixels in the landslide area, θ i is the slope value of the i-th pixel, in degrees;

[0087] Step 74: According to the landslide area contour, obtain the coordinates (x i ,y i ), and then calculate the projection area of ​​the landslide area on the horizontal plane. The calculation formula is as follows:

[0088]

[0089] Among them, (x i ,y i ) is the plane coordinate of the i-th vertex of the polygon, and n is the number of vertices of the polygon;

[0090] The average slope is then used to calculate the actual area of ​​the landslide slope, that is, the area of ​​forest landslide collapse, and the calculation formula is as follows:

[0091]

[0092] Among them, A 平面is the calculated plane projection area, and θ is the average slope of the landslide area.

[0093] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for detecting forest landslide and collapse area based on UAV visual remote sensing, characterized in that: The following steps are involved: Step 1: System construction: Construct a forest landslide collapse area detection system based on UAV visual remote sensing, the forest landslide collapse area detection system includes a UAV, the UAV is equipped with a multispectral camera, and the multispectral camera is sequentially connected to an image preprocessing module, an image alignment module, a landslide area recognition module and a landslide area calculation module; Step 2: Data acquisition: The UAV is equipped with a multispectral camera to take remote sensing images of the forest landslide area, obtain remote sensing image data, and pass it to the image preprocessing module; Step 3: Image preprocessing: The image preprocessing module obtains the historical satellite image data of the forest landslide area, and performs preprocessing operations on the historical satellite image data and remote sensing image data to obtain standard remote sensing images and historical images, and pass them to the image alignment module; Step 4: Image alignment: The image alignment module extracts feature points from the remote sensing image and the historical image, matches the feature points of the two, and then calculates a transformation matrix based on the matched feature point pairs, and maps the remote sensing image and the historical image to the same spatial reference system through the transformation matrix, thereby completing the alignment of the images before and after the landslide; Step 5: Landslide area identification: the landslide area identification module identifies the landslide area by calculating the ratio vegetation index RVI of the remote sensing image and the historical image; Step 6: Image post-processing: The landslide area recognition module uses the corrosion and expansion operations of OpenCV to remove small noises and connect the scattered landslide areas to extract the contour of the landslide area; Step 7: Calculation of landslide area: The landslide area calculation module uses the GDAL tool to perform coordinate conversion, generate a digital elevation model DEM, calculate the average slope of the landslide area, and then combine the plane projection area and the average slope to calculate the actual area of ​​the landslide slope to obtain the forest landslide collapse area.

2. The method for detecting forest landslide and collapse area based on UAV visual remote sensing according to claim 1 is characterized in that: In step 2, a multi-spectral camera mounted on an unmanned aerial vehicle is used to take multiple overlapping remote sensing images of the forest landslide area, and the heading overlap rate and the lateral overlap rate are not less than 80%.

3. The method for detecting forest landslide and collapse area based on UAV visual remote sensing according to claim 1 is characterized in that: In step 3, the image preprocessing module performs preprocessing operations on the remote sensing image data and the historical satellite image data. The specific steps are as follows: Step 31: Radiation calibration: The image preprocessing module converts the remote sensing image data and historical satellite image data stored in the form of digital values ​​DN into the form of radiation brightness values ​​through a conversion formula. The conversion formula is as follows: L=(DN-DN offset )×Gain Where L is the radiant brightness value, DN is the raw digital value, and DN offset is the offset value of the sensor, Gain is the gain coefficient; Step 32: Image correction: The image preprocessing module performs processing operations including but not limited to geometric correction, image fusion, image cropping, and shadow removal on the remote sensing image data and historical satellite image data in the form of radiation brightness values, and then uses the FLAASH atmospheric correction model to correct the radiation reflectivity to the surface reflectivity.

4. The method for detecting forest landslide and collapse area based on UAV visual remote sensing according to claim 3 is characterized in that: The geometric correction includes coarse correction, fine geometric correction and orthorectification.

5. The method for detecting forest landslide and collapse area based on UAV visual remote sensing according to claim 1, characterized in that: In step 4, the image alignment module uses the SIFT scale-invariant feature transformation method to detect feature points in the remote sensing image and the historical image, and then matches the feature points of the remote sensing image and the historical image through a nearest neighbor search algorithm to obtain a set of feature point pairs.

6. The method for detecting forest landslide and collapse area based on UAV visual remote sensing according to claim 1, characterized in that: In step 5, the landslide area identification module uses the cross-platform computer vision and machine learning software library OpenCV to read the remote sensing image and extract the data of the red light band and the near infrared band; the calculation expression of the ratio vegetation index RVI of the remote sensing image is as follows: Among them, I RED is the red light band data value of the corresponding pixel in the remote sensing image, I NIR It is the near-infrared band data value of the corresponding pixel in the remote sensing image.

7. The method for detecting forest landslide and collapse area based on UAV visual remote sensing according to claim 1, characterized in that: In step 6, the landslide area identification module uses the erosion function in OpenCV to perform erosion and expansion processing on the identified landslide area. The specific steps are as follows: Step 61: Select a 3X3 erosion convolution kernel and use the cv2.erode() function in OpenCV to perform an erosion operation on the landslide area image to obtain an eroded image; Step 62: Select a 3X3 dilated convolution kernel and use the cv2.dilate() function in OpenCV to dilate the eroded image to obtain a dilated image. Step 63: Use the contour extraction function cv2.findContours() in OpenCV to extract the contour of the expanded image and obtain the contour of the landslide area.

8. The method for detecting forest landslide and collapse area based on UAV visual remote sensing according to claim 1 or 7, characterized in that: In step 7, the landslide area calculation module calculates the forest landslide collapse area, and the specific steps are as follows: Step 71: The landslide area calculation module uses the `gdaltransform` tool in the open source raster spatial data conversion library GDAL to convert the pixel coordinates in the remote sensing image into geographic coordinates; uses the `ogr2ogr` tool in the open source raster spatial data conversion library GDAL to save the converted landslide boundary geographic coordinates into a vector format; Step 72: Use the 3D reconstruction open source software library OpenMVG to perform camera pose estimation and sparse point cloud generation. After the sparse point cloud is generated, further optimize the reconstruction model and generate a dense point cloud, and then use the GDAL tool to convert the dense point cloud into a digital elevation model DEM; each point of the dense point cloud contains X, Y and Z coordinates, and use the `gdal_grid` command in GDAL to interpolate each point of the dense point cloud to generate regular DEM grid data; Step 73: According to the DEM grid data, the slope is calculated using the gdaldem command in the GDAL tool; a mask is created to extract the slope data of the landslide area in the remote sensing image, and after all the slope values ​​in the landslide area are extracted, the average value of all the slope values ​​is calculated. The slope average value calculation formula is as follows: Where N is the total number of pixels in the landslide area, θ i is the slope value of the i-th pixel, in degrees; Step 74: According to the landslide area contour, obtain the coordinates (x i ,y i ), and then calculate the projection area of ​​the landslide area on the horizontal plane. The calculation formula is as follows: Among them, (x i ,y i ) is the plane coordinate of the i-th vertex of the polygon, and n is the number of vertices of the polygon; The average slope is then used to calculate the actual area of ​​the landslide slope, that is, the area of ​​forest landslide collapse, and the calculation formula is as follows: Among them, A 平面 is the calculated plane projection area, and θ is the average slope of the landslide area.

Images