A method, device, equipment, and medium for monitoring wind power ecology in desert steppe areas.

Abstract

This invention provides a method, device, equipment, and medium for monitoring the ecological environment of wind power in desert steppe areas. The method includes: acquiring remote sensing images of the target spatial area taken by satellite, aerial images taken by drones, and sensor information collected by ground sensors; obtaining a first standard raster map based on the aerial images; obtaining a second standard raster map based on the remote sensing images; inputting the first and second standard raster maps into a target network model for prediction to obtain a vegetation cover index (VC) and a desertification index (SDI); and determining the ecological level of wind power projects in desert steppe areas based on sensor information, VC, and SDI. This invention obtains the VC and SDI from the first and second standard raster maps, and, combined with sensor information on the ecological environment of wind power in desert steppe areas, achieves precise monitoring of the ecological safety of wind power in these areas.

Description

Technical Field

[0001] This invention relates to the field of ecological monitoring technology, and also to a method, device, equipment and medium for monitoring wind power ecology in desert grassland areas. Background Technology

[0002] In the ecological environment of wind power areas in desert grassland regions, there are problems such as fragile desert grassland ecology, frequent wind erosion, and sparse vegetation. However, most current ecological monitoring projects are not very targeted to the ecology of wind power in desert grassland regions, and the early warning mechanism is vague in level and the response measures are generalized, which makes it impossible to accurately identify the unique ecological security risks of desert grassland. This can easily lead to irreversible ecological damage due to untimely risk management. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a method, device, equipment and medium for monitoring wind power ecology in desert steppe areas, so as to solve the problem that the monitoring of wind power ecology in desert steppe areas is not targeted enough and cannot accurately identify the unique ecological security risks of desert steppe.

[0004] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: A method for monitoring the wind power ecology in desert steppe areas, comprising: Acquire remote sensing images of the target space region taken by satellites, aerial images taken by drones, and sensor information collected by ground sensors; The aerial image is preprocessed to obtain a first standard raster image; The remote sensing image is preprocessed to obtain a second standard raster image; the geographic coordinates of the first standard raster image and the second standard raster image are consistent. The first standard raster map and the second standard raster map are input into the target network model for prediction to obtain the vegetation cover index (VC) and the desertification index (SDI). The ecological level of wind power projects in desert grassland areas is determined based on the sensor information, vegetation cover index (VC), and desertification index (SDI).

[0005] Optionally, the aerial image is preprocessed to obtain a first standard raster image, including: Acquire the spatial location information and shooting posture information of the aerial image; The aerial images are stitched together based on the spatial location information and shooting posture information to obtain a first intermediate aerial image; Determine the point cloud image based on the first intermediate aerial image; Each point in the point cloud image is mosaicked according to its spatial location to obtain a second intermediate aerial image; Determine the target coordinate system of the second intermediate aerial image; The first standard raster image is determined based on the target coordinate system and the second intermediate aerial image.

[0006] Optionally, the remote sensing image is preprocessed to obtain a second standard raster image, including: The remote sensing image is subjected to geometric reprojection transformation to adjust the coordinate system of the remote sensing image to the target coordinate system, thereby obtaining an intermediate remote sensing image; The intermediate remote sensing image is cropped based on the spatial extent of the aerial image to obtain a second standard raster image.

[0007] Optionally, the first standard raster map and the second standard raster map are input into the target network model for prediction to obtain the vegetation cover index (VC) and the desertification index (SDI), including: The first standard raster image is input into the first network model of the target network model for local feature extraction to obtain the first feature; The second standard raster image is input into the first network model of the target network model for overall feature extraction to obtain the second feature; The first feature is input into the second network model of the target network model for prediction to obtain the visible vegetation index (VDVI). The second feature is input into the third network model of the target network model for prediction to obtain the Normalized Difference Vegetation Index (NDVI) and the Bare Soil Index (SI). The vegetation cover index VC is determined based on the normalized difference vegetation index NDVI and the visible light vegetation index VDVI. The land desertification index SDI is determined based on the vegetation cover index VC and the bare soil index SI.

[0008] Optionally, the ecological level of the wind power project in the desert steppe area is determined based on the sensor information, vegetation cover index (VC), and desertification index (SDI), including: Adjust the preset parameters based on the sensor information; The windbreak and sand-fixing function index WSI was determined based on the vegetation cover index (VC) and the desertification index (SDI). The windbreak and sand-fixing functional index (WSI) is compared with the preset parameters, and the ecological level of wind power in desert grassland areas is determined based on the comparison results.

[0009] Optionally, the sensing information includes wind speed, soil moisture, and wind erosion in the target space area; Adjusting preset parameters based on the sensor information includes: The first deviation value is determined based on wind speed, wind erosion, historical benchmark values ​​of wind speed and wind erosion. Determine the second deviation value between the soil moisture and the historical soil moisture baseline; The preset parameters are adjusted based on the first deviation value and the second deviation value.

[0010] Optionally, adjusting the preset parameter based on the first deviation value and the second deviation value includes: A first preset value is determined based on the first deviation value; Determine the second preset value based on the second deviation value; A third preset value is determined based on the first deviation value and the second deviation value; The preset parameters are determined based on the first preset value, the second preset value, and the third preset value.

[0011] A monitoring device for wind power ecology in desert grassland areas, comprising: The acquisition module is used to acquire remote sensing images of the target space area taken by satellites, aerial images taken by drones, and sensor information collected by ground sensors. The processing module is used to preprocess the aerial images to obtain a first standard raster image; preprocess the remote sensing images to obtain a second standard raster image; the geographic coordinates of the first standard raster image and the second standard raster image are consistent; input the first standard raster image and the second standard raster image into the target network model for prediction to obtain the vegetation cover index (VC) and the desertification index (SDI); and determine the ecological level of the wind power project in the desert grassland area based on the sensing information, the vegetation cover index (VC), and the desertification index (SDI).

[0012] The present invention also provides a computing device, comprising: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to perform the methods described above.

[0013] The present invention also provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores a program that, when executed by a processor, implements the above-described method.

[0014] The above-described solution of the present invention has at least the following beneficial effects: The above-described scheme of the present invention acquires remote sensing images of the target spatial area taken by satellite, aerial images taken by UAVs, and sensor information collected by ground sensors; preprocesses the aerial images to obtain a first standard raster map; preprocesses the remote sensing images to obtain a second standard raster map; the geographical coordinates of the first and second standard raster maps are consistent; the first and second standard raster maps are input into a target network model for prediction to obtain a vegetation cover index (VC) and a desertification index (SDI); and the ecological level of wind power projects in desert grassland areas is determined based on the sensor information, VC, and SDI. The present invention obtains the VC and SDI from the first and second standard raster maps, and, in conjunction with the sensor information on the ecological status of wind power in desert grassland areas, achieves accurate monitoring of the ecological safety of wind power in desert grassland areas. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating a method for monitoring wind power ecology in desert grassland areas according to the present invention. Figure 2 This is a schematic diagram of the structure of a monitoring device for wind power ecology in desert grassland areas according to the present invention. Detailed Implementation

[0016] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0017] like Figure 1 As shown, an embodiment of the present invention proposes a method for monitoring the wind power ecology in desert grassland areas, comprising: Step 11: Acquire remote sensing images of the target space area taken by satellites, aerial images taken by drones, and sensor information collected by ground sensors; Step 12: Preprocess the aerial image to obtain a first standard raster image; Step 13: Preprocess the remote sensing image to obtain a second standard raster image; the geographic coordinates of the first standard raster image and the second standard raster image are consistent; Step 14: Input the first standard raster map and the second standard raster map into the target network model for prediction to obtain the vegetation cover index VC and the land desertification index SDI; Step 15: Determine the ecological level of the wind power project in the desert grassland area based on the sensor information, vegetation cover index (VC), and desertification index (SDI).

[0018] In this embodiment, wind power project facilities, such as photovoltaic panels, are often located in desert steppe areas. To specifically monitor the ecological management of desert steppe areas with wind power equipment, this invention not only considers images taken at different altitudes but also incorporates the vegetation cover index (VC) and desertification index (SDI) (the VC and SDI of desert steppe areas with wind power equipment differ from those of ordinary desert steppe areas). Combined with sensor information on the wind power ecology in desert steppe areas, this enables precise monitoring of the ecological safety of wind power in these areas.

[0019] For step 11, the target spatial area refers to the desert grassland area where wind power equipment is installed. Remote sensing images are captured from a high distance, while drone images are captured from a low distance. The aerial images of the target spatial area taken by the drone are multiple photographs, which are stitched together to present a complete picture of the target spatial area. The sensing information includes wind speed, soil moisture, and wind erosion in the target spatial area. Generally, drone images have a resolution of 0.1m, which can well show local details. Remote sensing images have a resolution of 10m, which can well show global features.

[0020] Among them, remote sensing images taken by satellites: high-resolution optical satellites with a resolution of less than 1m can be selected, and the combination of near-infrared and red light bands is optimized; data are collected monthly during the operation period, weekly during the construction period, and every 15 days during the winter and spring windy and dusty seasons, and invalid data is automatically removed and preliminary geometric correction is completed.

[0021] Aerial images captured by drones: Industrial-grade drones equipped with high-definition cameras and LiDAR are specially designed for dust protection, wind resistance (greater than level 7), and sun exposure resistance; centimeter-level aerial photography is carried out on core disturbance areas such as wind turbine foundations and construction access roads, and ecologically sensitive points are monitored at high frequency. Three-dimensional point cloud models are automatically generated to accurately capture micro-disturbances on the ground, providing data support for the refined quantification of the three major indicators in the core area.

[0022] Ground-based sensors collect sensor information: multi-element integrated sensors (monitoring wind speed, soil moisture, wind erosion, etc.), and are designed to withstand dust and extreme temperatures. The device features a low-power design and adopts a dual power supply mode of solar energy and lithium battery. It is linearly deployed along the wind turbine array and ecological protection red line, collecting data once per hour to provide real-time data related to wind erosion and vegetation growth for the calculation of the sand fixation function index.

[0023] In this embodiment, after data collection, data denoising, spatiotemporal alignment, and format standardization are also performed; 5G+BeiDou dual-mode transmission is adopted, with 5G as the primary mode and BeiDou as the backup, to achieve real-time data backhaul.

[0024] In some alternative implementations, step 12, preprocessing the aerial image to obtain a first standard raster image, includes: Step 121: Obtain the spatial location information and shooting posture information of the aerial image; Step 122: The aerial images are stitched together according to the spatial location information and shooting posture information to obtain a first intermediate aerial image; Step 123: Determine the point cloud image based on the first intermediate aerial image; Step 124: Mosaic each point in the point cloud image according to its spatial location to obtain the second intermediate aerial image; Step 125: Determine the target coordinate system of the second intermediate aerial image; Step 126: Determine the first standard raster image based on the target coordinate system and the second intermediate aerial image.

[0025] For step 121, the drone is equipped with positioning and attitude data to collect each aerial photograph. Spatial position information includes longitude (Lon), latitude (Lat), and shooting altitude (H), while shooting attitude information includes roll angle. Pitch angle Yaw angle Each aerial image can be represented as .

[0026] For step 122, adjacent aerial images are determined based on spatial location information. SIFT feature points are extracted from the adjacent aerial images (i.e., corresponding points in the adjacent aerial images are determined). The transformation matrix that minimizes the pixel difference in the overlapping area of ​​the two images is calculated. ; in, and For adjacent aerial images, It is a homography matrix. Indicates the image The points are projected onto the H matrix through transformation. The position of n, where n is the number of feature points. express The i-th feature point, express and The homography matrix with the minimum difference is the transformation matrix that minimizes the pixel difference in the overlapping area of ​​two images. This matrix includes the feature points that match between two adjacent images.

[0027] Then, according to Determine the rotation matrix for each image. Translation vector Where s represents the s-th drone photo, This represents the j-th point in three-dimensional space (grass, stone, wind turbine on the ground). This indicates the total number of photos taken by the drone. Represents the total number of points in space. Represents the rotation matrix for each photo. Translation vector The set, This represents the actual pixel coordinates (measured values) of point j as seen in photo s. This represents the rotation matrix (i.e., pitch, roll, and yaw attitude matrix) of the s-th photo. This represents the 3D coordinates of point j in the real world. Let represent the translation vector of the s-th photo (the drone's position at that time). This means transforming 3D points in the world into the camera coordinate system. This represents the camera projection function (i.e., projecting 3D points into 2D pixels).

[0028] After determining the corresponding points of adjacent photos and correcting the adjacent photos, the rotation matrix is ​​used as the basis for the calculation. Translation vector Correct the position and orientation of all drone photos to ensure the entire stitched image is free of misalignment and distortion. Then, stitch the corrected photos together with corresponding points from adjacent photos to combine the drone's aerial images. By stitching the photos together, the first intermediate aerial image is obtained, in which... This represents the final merged pixel values, i.e., the pixel values ​​in the first intermediate aerial image. This represents the weighting coefficient (values ​​from 0 to 1), the closer to image A... The closer it is to 1, The table shows the pixels at this location in the corrected image A. This represents the pixel at that location in the corrected image B. Images A and B are adjacent images.

[0029] For steps 123, according to The 3D world coordinates of ground points in the sparse but precise sparse point cloud (determining the overall spatial structure) corresponding to feature points in the first intermediate aerial image are obtained, where, This represents the coordinates of the feature points that are matched between two adjacent images used to stitch together the first intermediate aerial image. This represents the camera intrinsic parameter matrix (focal length, principal point, etc.). This represents the rotation matrix (i.e., pitch, roll, and yaw attitude matrix) of the s-th photo. Let represent the translation vector of the s-th photo (the drone's position at that time). This represents the 3D world coordinates of the ground points in the sparse but precise point cloud corresponding to the feature points in each image stitched together from the first intermediate aerial image.

[0030] according to Determine whether the point clouds of each image stitched together from the first intermediate aerial image are from the same ground point. This represents the pixels corresponding to the 3D world coordinates of the ground points in the first image of the first intermediate aerial image stitched together. This represents the pixel coordinates of the ground points in the second image, which is stitched together from the first intermediate aerial image, corresponding to the 3D world coordinates. These are actually pixels from the same pixel. Represents pixels grayscale value, Represents pixels grayscale value, Indicates The average pixel value within the centered window. Indicates The average pixel value within the centered window. This represents the normalized cross-correlation coefficient. The closer this coefficient is to 1, the more likely that the two pixels are the same ground point.

[0031] Sequentially determine whether the point cloud of each image stitched together from the first intermediate aerial image is the same ground point, and then... Points with values ​​greater than 0.9 are considered to be the same ground point, based on the aforementioned camera intrinsic matrix K and extrinsic parameters. , The triangulation method is used to perform spatial forward intersection of pixels with the same name to obtain the coordinates of the corresponding three-dimensional points on the ground. The point cloud of each image of the first intermediate aerial image is traversed and the above calculation is repeated to finally obtain a point cloud image covering the entire target area.

[0032] For step 124, based on the three-dimensional world coordinates of each point in the point cloud image... The target area is divided into regular grids according to a preset ground resolution. The coordinates of the grid center point are: ; in, These are the minimum X-coordinate and minimum Y-coordinate of each point in the point cloud image; p and q are the preset resolutions of the grid cells in the X and Y directions, respectively, and the grid row and column indices are p and q.

[0033] For each grid (p, q), according to Calculate the pixel values ​​of this grid, where, This represents the pixel value at grid (p, q) in the second intermediate aerial image. This represents the number of point cloud points that fall within the current grid's neighborhood. Represented as the RGB pixel value of the k-th point in the point cloud image. Represented as the k-th point cloud point and the current grid center point ( The planar distance between ).

[0034] After assigning values ​​to all grid cells in sequence, a two-dimensional orthophoto image is formed that covers the entire target area, has no holes, and has a consistent spatial position. This is the second intermediate aerial image.

[0035] For step 125, the target coordinate system adopts the UAV's WGS84 latitude and longitude, projected into UTM plane coordinates to ensure accurate overlay with the subsequent satellite image. Using the 3D world coordinate system corresponding to the point cloud image in step 123 as a reference, the target coordinate system is determined to be a WGS84 UTM projection plane rectangular coordinate system, whose coordinate system parameters satisfy: ; in, The longitude and latitude spatial location information collected by the UAV in step 121, The altitude at which the drone takes pictures in step 121. This is the projection transformation function from WGS84 latitude and longitude to UTM plane coordinates. These are the three-dimensional coordinates in the target coordinate system.

[0036] Simultaneously, using the bounding rectangle of the point cloud coverage area as the spatial range, the origin, coordinate axis direction, and scale of the target coordinate system are determined, ensuring that the position of each pixel in the second intermediate aerial image is aligned with the planar coordinates under this target coordinate system. Correspondingly, this ensures that the second standard raster map, which is subsequently preprocessed from remote sensing images, has a consistent geographic coordinate system reference.

[0037] For step 126, all pixels of the second intermediate aerial image are mapped to the target coordinate system through coordinate transformation. Next, resampling and geocoding are performed at a uniform resolution to obtain the first standard raster map. The transformation relationship is as follows: ,in, This refers to the pixel row and column number of the second intermediate aerial image. These are the planar coordinates of the corresponding ground points in the target coordinate system. This is a mapping function between pixel coordinates and geographic coordinates in the target coordinate system.

[0038] During the resampling process, through Calculate the standard raster pixel values, where, For the first standard raster map in Pixel value at that location, The values ​​of the neighboring pixels in the second intermediate aerial image. The bilinear interpolation weights are determined based on pixel distance.

[0039] After geographic coordinate binding and resampling, the output is a first standard raster map with a unified target coordinate system, fixed resolution, and strict spatial alignment.

[0040] In some alternative implementations, step 13, preprocessing the remote sensing image to obtain a second standard raster image, includes: Step 131: Perform a geometric reprojection transformation on the remote sensing image to adjust the coordinate system of the remote sensing image to the target coordinate system, thereby obtaining an intermediate remote sensing image; Step 132: Crop the intermediate remote sensing image according to the spatial range of the aerial image to obtain a second standard raster image.

[0041] In this embodiment, the latitude and longitude coordinate system of the original remote sensing image is unified to the WGS84 UTM target coordinate system determined in step 125 through projection transformation. The transformation formula is as follows: ; in, These are the latitude and longitude coordinates corresponding to the pixels in the original remote sensing image. Let be the projection transformation function from WGS84 latitude and longitude to UTM plane coordinates, which is consistent with the projection function in step 125. The coordinates of the remote sensing image pixels are in the target coordinate system.

[0042] Establish the original pixel coordinates of the remote sensing image Coordinates with the target coordinate system Mapping relationship between them: ; in, The starting coordinates of the top left corner of the remote sensing image in the target coordinate system. The resolution is the original remote sensing image (10m). These are the image rotation and azimuth correction coefficients, used to correct geometric distortions in remote sensing images.

[0043] The coordinates of the remote sensing image are remapped pixel by pixel, and bilinear interpolation is used to complete grayscale resampling to obtain an intermediate remote sensing image that is completely consistent with the coordinate system of the first standard raster image.

[0044] Then, based on the point cloud coverage area in step 124, determine the outer rectangular boundary of the target region in the target coordinate system: ; in, Let the minimum and maximum coordinates of the point cloud coverage area in the X direction in the target coordinate system be denoted as . These are the minimum and maximum coordinates of the point cloud coverage area in the Y direction within the target coordinate system.

[0045] Based on the aforementioned bounding rectangle, the intermediate remote sensing image is spatially cropped, retaining only pixels falling within the rectangle's area. The cropping determination rule is as follows: ; in, Let (u,v) be the pixel value at pixel (u,v) in the intermediate remote sensing image. This refers to the coordinates of the pixel in the target coordinate system. These are the valid pixels retained after cropping.

[0046] The cropped image is geocoded and rasterized to obtain a second standard raster image, which has the same geographic coordinate system and spatial range as the first standard raster image, but a different resolution, to meet the multi-scale feature input requirements of the subsequent network model.

[0047] In some optional implementations, step 14 involves inputting the first standard raster map and the second standard raster map into the target network model for prediction to obtain the vegetation cover index (VC) and the desertification index (SDI), including: Step 141: Input the first standard grid image into the first network model of the target network model to extract local features and obtain the first feature; Step 142: Input the second standard grid image into the first network model of the target network model for overall feature extraction to obtain the second feature; Step 143: Input the first feature into the second network model of the target network model for prediction to obtain the visible vegetation index (VDVI). Step 144: Input the second feature into the third network model of the target network model for prediction to obtain the Normalized Difference Vegetation Index (NDVI) and the Bare Soil Index (SI). Step 145: Determine the vegetation cover index VC based on the normalized difference vegetation index NDVI and the visible light vegetation index VDVI; Step 146: Determine the land desertification index SDI based on the vegetation cover index VC and the bare soil index SI.

[0048] For steps 141 and 142, in this embodiment, the first network model is used to extract feature values, and the extracted feature values ​​are input as the second network model of the classifier to obtain the visible vegetation index (VDVI), normalized difference vegetation index (NDVI), and bare soil index (SI).

[0049] The first network model employs a multi-layer convolutional neural network to extract spatial features, performing local and global feature extraction on the first and second standard grid images, respectively. Specifically, the first and second standard grid images are input into the first network model, and the features of the first and second standard grid images are extracted through multiple convolutional layers in the first network model. The first network model includes multiple single-layer convolutional layers and multiple pooling layers. The formula for calculating a single convolutional layer is: ,in, Features of the first standard raster image or features of the second standard raster image. The convolution kernel weight matrix is... For the input values ​​of a single convolutional layer, This is for kernel bias.

[0050] Then, the features of the first and second standard raster images are downsampled through a pooling layer to obtain first and second features of the same size. The pooling layer calculation formula is as follows: ,in, This is the output of the pooling layer (i.e., the first feature or the second feature). This is the output of the previous single-layer convolutional layer. This represents the starting row number of the current pooling window on the first or second feature map. This is the starting column number of the current pooling window on the first or second feature map. The height of the pooled window in the row direction (the window size in the vertical direction). This represents the width of the pooled window in the column direction (the horizontal window size). To define the coverage area of ​​the pooling window in the row direction, starting from the first row... Arrive at the OK, To define the coverage of the pooling window in the column direction, starting from the first... Listed to number List.

[0051] For steps 143 and 144, the input to the second network model is the first feature. The output is the Visible Light Vegetation Index (VDVI). When training the second network model, the input sample set is... ,in This represents the i-th sample with the first feature. This represents the true label of the i-th sample. , i=1,2,…,m.

[0052] Before training the second network model, define the number T of decision trees in the second network model and the maximum depth of each tree. Minimum number of samples for node splitting Minimum number of samples in a leaf node The number of features L during node splitting is preferably T=200. =15, =10, =5.

[0053] Construction of the second network model: 1. From the original dataset The training set for the tree is generated by sampling with replacement. .

[0054] 2. Clearly define the target node for splitting to minimize the error of the Visible Difference Vegetation Index (VDVI). ,in, This represents the mean square error of the visible vegetation index (VDVI). This represents the true visible light vegetation index. The visible light vegetation index is the result of the prediction.

[0055] 3. Find the The feature that decreases the most and the threshold are used to split the left and right child nodes. This continues until the maximum depth is reached or the number of samples is insufficient.

[0056] When using the second network model, the average value of the Visible Light Vegetation Index (VDVI) generated by the T trees is output.

[0057] Among them, according to Determine the feature number L when a node splits, where, , where n is the total number of dimensions of the first feature. represents the weight of the feature dimension of the (t-1)th tree, with values ​​ranging from 0.4 to 0.6. Let be the effective dimension of the final optimal splitting feature for the (t-1)th tree. Adjust the weights based on the fitting accuracy of the previous tree, with values ​​ranging from 0.4 to 0.6. Let t be the mean of the visible vegetation index (VDVI) of the previous tree, and t represent the current t-th decision tree.

[0058] The input to the third network model is the second feature. The output is a 2D continuous value. When training the third network model, the input sample set is ,in Let i represent the sample with the second feature. Let represent the true label (2-dimensional vector) of the i-th sample. , i=1,2,…,m.

[0059] Before training the third network model, define the number T of decision trees in the third network model and the maximum depth of each tree. Minimum number of samples for node splitting Minimum number of samples in a leaf node The number of features k when a node splits is preferably T=200. =15, =10, =5.

[0060] Construction of the third network model: 1. Generate the training set for the tree by sampling with replacement from the original dataset D. .

[0061] 2. Clearly define the target node for minimizing the errors of both the Normalized Difference Vegetation Index (NDVI) and the Bare Soil Index (SI). ,in, ,in, The goal is to split the smallest node. Normalized Difference Vegetation Index (NDVI) For the true normalized vegetation index, To predict the normalized vegetation index, This represents the mean square error of the bare soil index. To represent the true bare soil index, The predicted bare soil index.

[0062] 3. Find the The feature that decreases the most and the threshold are used to split the left and right child nodes. This continues until the maximum depth is reached or the number of samples is insufficient.

[0063] When using the third network model, the mean of the Normalized Difference Vegetation Index (NDVI) generated by T trees is output as the final Normalized Difference Vegetation Index (NDVI), and the mean of the Bare Soil Index (SI) generated by T trees is output as the Bare Soil Index (SI).

[0064] Among them, according to Determine the feature number k when a node splits, where, , where n is the total number of dimensions of the second feature. represents the weight of the feature dimension of the (t-1)th tree, with values ​​ranging from 0.4 to 0.6. Let be the effective dimension of the final optimal splitting feature for the (t-1)th tree. Adjust the weights based on the fitting accuracy of the previous tree, with values ​​ranging from 0.4 to 0.6. Let t be the mean of the joint prediction coefficients of NDVI and SI by the previous tree, and t represent the current t-th decision tree.

[0065] For step 145, according to Determine the vegetation cover index, among which , For fixed fusion weights Preferred , , Normalized Difference Vegetation Index (NDVI) The visible vegetation index.

[0066] For step 146, according to The land desertification index was determined, among which... This refers to the desertification index of the desert grassland wind power area; the higher the value, the more severe the desertification. The vegetation cover index, Contributes weight to bare soil desertification. Assuming the weight of vegetation in sand fixation, the preferred method is... , .

[0067] In some alternative implementations, step 15 involves basing the sensor information and vegetation cover index on the data. VC and the Soil Desertification Index (SDI) determine the ecological class of wind power projects in desert steppe areas, including: Step 151: Adjust the preset parameters according to the sensor information; Step 152: Determine the windbreak and sand-fixing function index WSI based on the vegetation cover index VC and the desertification index SDI; Step 153: Compare the windbreak and sand-fixing functional index (WSI) with the preset parameters, and determine the ecological level of wind power in the desert grassland area based on the comparison results.

[0068] In some alternative implementations, the sensing information includes wind speed, soil moisture, and wind erosion in the target spatial area; And, step 151, adjusting preset parameters according to the sensing information, includes: Step 1511: Determine the first deviation value based on wind speed, wind erosion, historical wind speed benchmark value, and historical wind erosion benchmark value; Step 1512: Determine the second deviation value between the soil moisture and the historical soil moisture baseline value; Step 1513: Adjust the preset parameters according to the first deviation value and the second deviation value.

[0069] In this embodiment, according to Determine the first deviation value, where, The first deviation value represents the overall degree of deviation between real-time wind speed and wind erosion relative to the historical baseline. The current wind speed is obtained through real-time monitoring of the target spatial area. The historical benchmark wind speed (normal optimal / steady-state reference wind speed) is the long-term statistical standard for the wind power area in this desert grassland. The current wind erosion amount is collected in real time by sensors in the target space area. This is the historical baseline value of wind erosion under the ecological stability of the region (the standard wind erosion value under normal conditions without exacerbating desertification).

[0070] according to Determine the second deviation value, where, The second deviation value characterizes the degree of water deficit deviation of real-time soil moisture relative to the historical baseline. The current soil moisture is collected in real time by ground sensors in the target spatial area. This is the historical baseline value of soil moisture determined under the conditions of vegetation sand fixation and ecological stability in the desert grassland wind power area.

[0071] In some optional implementations, step 1513, adjusting the preset parameter based on the first deviation value and the second deviation value, includes: Step 15131: Determine a first preset value based on the first deviation value; Step 15132: Determine the second preset value based on the second deviation value; Step 15133: Determine a third preset value based on the first deviation value and the second deviation value; Step 15134: Determine the preset parameters based on the first preset value, the second preset value, and the third preset value.

[0072] In this embodiment, according to Determine the first preset value, where, The first preset value (wind erosion risk correction item, the highest level). These are the original preset parameters. The weighting coefficient for the first deviation value ( (preferably 0.5%) This is the first deviation value.

[0073] In this embodiment, according to Determine the second preset value, where, The second preset value (soil moisture correction item, compared to) Low), The first preset value, The weighting coefficient for the second deviation value ( (preferably 0.4) This is the second deviation value.

[0074] In the embodiments, according to Determine the third preset value; in, This indicates the adjusted preset parameters. This represents the original preset parameters. This represents the weighting coefficient for the first deviation value. This represents the weighting coefficient for the second deviation value. Indicates the first deviation value, according to Determine the first deviation value, where, This indicates real-time monitoring of wind speed. This represents the historical baseline value of wind speed. This indicates real-time monitoring of wind erosion. This represents the historical baseline value for wind erosion. This represents the second deviation value. Determine the second deviation value, where, This represents the historical baseline value of soil moisture. This indicates real-time monitoring of soil moisture.

[0075] For step 152, according to The Windbreak and Sand Fixation Function Index (WSI) was determined, among which... , For fixed weighting coefficients, the preferred method is... , This is the vegetation cover index (a positive gain term for vegetation sand fixation). This is the land desertification index (a negative weakening term for desertification). The windbreak and sand fixation function index indicates that the higher the value, the stronger the region's windbreak and sand fixation capacity and the better its ecological stability; the lower the value, the higher the risk of desertification and the weaker the sand fixation capacity.

[0076] In some embodiments, , It can be calculated .

[0077] For step 153, if The ecological rating of wind power in the desert steppe area is blue (general). The WSI (Weather Safety Index) is within the safe range, and the vegetation cover and land desertification index both meet the specific thresholds for desert steppe, indicating no ecological risk. Only one indicator is close to the threshold (early warning). The ecological rating of wind power in the desert steppe area is yellow (relatively severe), with a slight decrease in WSI and a single core indicator (VC or SDI) exceeding the desert steppe-specific threshold, indicating initial damage to windbreak and sand-fixing functions, requiring targeted management; if The ecological level of wind power in the desert steppe area is orange (severe), with moderate damage to the Wind Sound Indicator (WSI). Both core indicators (VC+SDI) have exceeded the thresholds specific to desert steppes. The combined effects of desertification and vegetation damage have significantly reduced the windbreak and sand-fixing function, requiring urgent intervention. The ecological level of wind power in the desert grassland area is red (particularly serious), with severe WSI damage (core judgment threshold in the disclosure document), or irreversible ecological damage (such as large-scale permanent desertification, complete destruction of vegetation in the core area), and the core function of windbreak and sand fixation has failed, requiring the initiation of a comprehensive emergency repair project.

[0078] Among them, the abnormal distribution of the three core indicators and the distribution of ecological risks can be visualized through heat maps, 3D modeling, and risk point marking; early warning information is pushed to computer terminals, mobile APPs, and SMS platforms in real time, and the information includes risk level, precise location, abnormal data of core indicators, cause analysis and emergency response suggestions.

[0079] In the above embodiments of the present invention, the simulation prediction layer constructs three dynamic numerical simulation models specifically for desert steppes: a first network model, a second network model, and a third network model. These models respectively represent a dynamic simulation model of the desert steppe ecosystem, a simulation model of ecological disturbance caused by wind power projects, and a simulation model of the impact of migratory birds. This enables prediction of ecological security trends throughout the entire lifecycle of wind power projects, from construction to operation to recovery. The safety early warning layer establishes a four-level ecological security early warning mechanism (blue, yellow, orange, and red), sets early warning thresholds specific to desert steppes, and can automatically generate customized ecological risk response measures based on the early warning level. Early warning information supports multi-terminal push notifications and visualization on computers and mobile devices. Real-time acquisition of ecological data is achieved through a high-frequency, high-precision data collection system combining satellite, drones, and ground-based methods. Combined with the three dynamic numerical simulation models specific to desert steppes, the system can accurately capture dynamic ecological changes throughout the entire lifecycle of wind power projects, from construction to operation to recovery. The prediction error is reduced by more than 35% compared to existing general models, and the long-term cumulative impact of projects on windbreak and sand fixation functions can be quantified, providing a scientific basis for real-time adjustments to ecological protection measures.

[0080] like Figure 2 As shown, an embodiment of the present invention proposes a monitoring device 20 for wind power ecology in desert grassland areas, comprising: The acquisition module 21 is used to acquire remote sensing images of the target space area taken by satellites, aerial images taken by drones, and sensing information collected by ground sensors. Processing module 22 is used to preprocess the aerial image to obtain a first standard raster image; preprocess the remote sensing image to obtain a second standard raster image; the geographic coordinates of the first standard raster image and the second standard raster image are consistent; input the first standard raster image and the second standard raster image into the target network model for prediction to obtain the vegetation cover index (VC) and the desertification index (SDI); and determine the ecological level of the wind power project in the desert grassland area based on the sensing information, the vegetation cover index (VC), and the desertification index (SDI).

[0081] Optionally, the aerial image is preprocessed to obtain a first standard raster image, including: Acquire the spatial location information and shooting posture information of the aerial image; The aerial images are stitched together based on the spatial location information and shooting posture information to obtain a first intermediate aerial image; Determine the point cloud image based on the first intermediate aerial image; Each point in the point cloud image is mosaicked according to its spatial location to obtain a second intermediate aerial image; Determine the target coordinate system of the second intermediate aerial image; The first standard raster image is determined based on the target coordinate system and the second intermediate aerial image.

[0082] Optionally, the remote sensing image is preprocessed to obtain a second standard raster image, including: The remote sensing image is subjected to geometric reprojection transformation to adjust the coordinate system of the remote sensing image to the target coordinate system, thereby obtaining an intermediate remote sensing image; The intermediate remote sensing image is cropped based on the spatial extent of the aerial image to obtain a second standard raster image.

[0083] Optionally, the first standard raster map and the second standard raster map are input into the target network model for prediction to obtain the vegetation cover index (VC) and the desertification index (SDI), including: The first standard raster image is input into the first network model of the target network model for local feature extraction to obtain the first feature; The second standard raster image is input into the first network model of the target network model for overall feature extraction to obtain the second feature; The first feature is input into the second network model of the target network model for prediction to obtain the visible vegetation index (VDVI). The second feature is input into the third network model of the target network model for prediction to obtain the Normalized Difference Vegetation Index (NDVI) and the Bare Soil Index (SI). The vegetation cover index VC is determined based on the normalized difference vegetation index NDVI and the visible light vegetation index VDVI. The land desertification index SDI is determined based on the vegetation cover index VC and the bare soil index SI.

[0084] Optionally, the ecological level of the wind power project in the desert steppe area is determined based on the sensor information, vegetation cover index (VC), and desertification index (SDI), including: Adjust the preset parameters based on the sensor information; The windbreak and sand-fixing function index WSI was determined based on the vegetation cover index (VC) and the desertification index (SDI). The windbreak and sand-fixing functional index (WSI) is compared with the preset parameters, and the ecological level of wind power in desert grassland areas is determined based on the comparison results.

[0085] Optionally, the sensing information includes wind speed, soil moisture, and wind erosion in the target space area; Adjusting preset parameters based on the sensor information includes: The first deviation value is determined based on wind speed, wind erosion, historical benchmark values ​​of wind speed and wind erosion. Determine the second deviation value between the soil moisture and the historical soil moisture baseline; The preset parameters are adjusted based on the first deviation value and the second deviation value.

[0086] Optionally, adjusting the preset parameter based on the first deviation value and the second deviation value includes: A first preset value is determined based on the first deviation value; Determine the second preset value based on the second deviation value; A third preset value is determined based on the first deviation value and the second deviation value; The preset parameters are determined based on the first preset value, the second preset value, and the third preset value.

[0087] It should be noted that this device is the same as the method described above. All implementations in the above method embodiments are applicable to the embodiments of this device and can achieve the same technical effect.

[0088] Embodiments of the present invention also provide a computing device, including: one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the above-described method. All implementations in the above method embodiments are applicable to the embodiments of this computing device and can achieve the same technical effects.

[0089] In another aspect, the present invention also provides a computer-readable storage medium storing a program that, when executed by a processor, implements the above-described method. All implementations in the above method embodiments are applicable to the embodiments of this computer-readable storage medium and can achieve the same technical effects.

[0090] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for monitoring the wind power ecology in desert grassland areas, characterized in that, include: Acquire remote sensing images of the target space region taken by satellites, aerial images taken by drones, and sensor information collected by ground sensors; The aerial image is preprocessed to obtain a first standard raster image; The remote sensing image is preprocessed to obtain a second standard raster image; The geographic coordinates of the first standard raster map and the second standard raster map are consistent; The first standard raster map and the second standard raster map are input into the target network model for prediction to obtain the vegetation cover index (VC) and the desertification index (SDI). The ecological level of wind power projects in desert grassland areas is determined based on the sensor information, vegetation cover index (VC), and desertification index (SDI).

2. The monitoring method for wind power ecology in desert steppe areas according to claim 1, characterized in that, The aerial image is preprocessed to obtain a first standard raster image, including: Acquire the spatial location information and shooting posture information of the aerial image; The aerial images are stitched together based on the spatial location information and shooting posture information to obtain a first intermediate aerial image; Determine the point cloud image based on the first intermediate aerial image; Each point in the point cloud image is mosaicked according to its spatial location to obtain a second intermediate aerial image; Determine the target coordinate system of the second intermediate aerial image; The first standard raster image is determined based on the target coordinate system and the second intermediate aerial image.

3. The monitoring method for wind power ecology in desert steppe areas according to claim 2, characterized in that, The remote sensing image is preprocessed to obtain a second standard raster image, including: The remote sensing image is subjected to geometric reprojection transformation to adjust the coordinate system of the remote sensing image to the target coordinate system, thereby obtaining an intermediate remote sensing image; The intermediate remote sensing image is cropped based on the spatial extent of the aerial image to obtain a second standard raster image.

4. The monitoring method for wind power ecology in desert grassland areas according to claim 1, characterized in that, The first and second standard raster maps are input into the target network model for prediction to obtain the vegetation cover index (VC) and the desertification index (SDI), including: The first standard raster image is input into the first network model of the target network model for local feature extraction to obtain the first feature; The second standard raster image is input into the first network model of the target network model for overall feature extraction to obtain the second feature; The first feature is input into the second network model of the target network model for prediction to obtain the visible vegetation index (VDVI). The second feature is input into the third network model of the target network model for prediction to obtain the Normalized Difference Vegetation Index (NDVI) and the Bare Soil Index (SI). The vegetation cover index VC is determined based on the normalized difference vegetation index NDVI and the visible light vegetation index VDVI. The land desertification index SDI is determined based on the vegetation cover index VC and the bare soil index SI.

5. The monitoring method for wind power ecology in desert steppe areas according to claim 1, characterized in that, The ecological level of wind power projects in desert grassland areas is determined based on the aforementioned sensor information, vegetation cover index (VC), and desertification index (SDI), including: Adjust the preset parameters based on the sensor information; The windbreak and sand-fixing function index WSI was determined based on the vegetation cover index (VC) and the desertification index (SDI). The windbreak and sand-fixing functional index (WSI) is compared with the preset parameters, and the ecological level of wind power in desert grassland areas is determined based on the comparison results.

6. The monitoring method for wind power ecology in desert steppe areas according to claim 5, characterized in that, The sensing information includes wind speed, soil moisture, and wind erosion in the target space area; Adjusting preset parameters based on the sensor information includes: The first deviation value is determined based on wind speed, wind erosion, historical benchmark values ​​of wind speed and wind erosion. Determine the second deviation value between the soil moisture and the historical soil moisture baseline; The preset parameters are adjusted based on the first deviation value and the second deviation value.

7. The monitoring method for wind power ecology in desert steppe areas according to claim 6, characterized in that, Adjusting the preset parameters based on the first deviation value and the second deviation value includes: A first preset value is determined based on the first deviation value; Determine the second preset value based on the second deviation value; A third preset value is determined based on the first deviation value and the second deviation value; The preset parameters are determined based on the first preset value, the second preset value, and the third preset value.

8. A monitoring device for wind power ecology in desert grassland areas, characterized in that, include: The acquisition module is used to acquire remote sensing images of the target space area taken by satellites, aerial images taken by drones, and sensor information collected by ground sensors. The processing module is used to preprocess the aerial image to obtain a first standard raster image; The remote sensing image is preprocessed to obtain a second standard raster image; the geographic coordinates of the first standard raster image and the second standard raster image are consistent; the first standard raster image and the second standard raster image are input into the target network model for prediction to obtain the vegetation cover index (VC) and the desertification index (SDI); the ecological level of the wind power project in the desert grassland area is determined based on the sensing information, the vegetation cover index (VC), and the desertification index (SDI).

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images