An unmanned aerial vehicle cooperative remote sensing technology-based intelligent soil erosion monitoring method and system

Abstract

The present application relates to the technical field of image processing, and more particularly to a soil and water loss intelligent monitoring method and system of unmanned aerial vehicle cooperative remote sensing technology. The method comprises the following steps: acquiring geographic data of a target area and dividing monitoring grids; determining a task allocation sequence of a group of unmanned aerial vehicles according to each monitoring grid; controlling the group of unmanned aerial vehicles to enter the corresponding monitoring grid according to the task allocation sequence to collect remote sensing image data; splicing and registering the remote sensing image data collected by different unmanned aerial vehicles and classifying and identifying the surface type; calculating the surface disturbance intensity parameters of each monitoring grid according to the surface type change and the texture feature change amplitude of the same monitoring grid in adjacent image frames in the multi-source remote sensing fusion data; and weighting correcting the surface disturbance intensity parameters of each monitoring grid to determine the dynamic evaluation results of soil and water loss of each monitoring grid. The present application can improve the recognition ability of the soil and water loss process and improve the response speed and early warning accuracy of soil and water loss monitoring.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to an intelligent monitoring method and system for soil erosion using UAV-assisted remote sensing technology. Background Technology

[0002] Soil erosion, a significant factor restricting regional ecological security and sustainable land resource utilization, is widespread in mountainous, hilly, and wind-eroded areas. Its formation is influenced by the coupled effects of multiple factors, including rainfall intensity, topographic slope, soil structure, and vegetation cover, exhibiting significant spatiotemporal dynamics and nonlinear evolution characteristics. Traditional soil erosion monitoring methods primarily rely on ground-based runoff plot observations, manual sampling measurements, and data from fixed-point hydrological monitoring stations. While these methods offer a certain level of accuracy, they suffer from limitations such as limited monitoring range, long data acquisition cycles, insufficient spatial representativeness, and delayed response to sudden erosion events, making them unsuitable for large-scale, dynamic soil erosion monitoring needs.

[0003] With the gradual development of multi-UAV collaborative remote sensing technology, the rapid coverage of large areas and multi-view data acquisition through the collaborative operation of multiple UAVs have improved monitoring efficiency to some extent. However, in existing UAV remote sensing technologies, the multi-UAV collaborative process often relies on preset flight paths or simple task allocation strategies, lacking an adaptive collaborative mechanism for the dynamic characteristics of soil erosion. At the same time, the data acquired by different UAVs differ in time reference, spatial scale, and sensor type, making it difficult to fuse multi-source data and form unified, high-precision soil erosion monitoring results. Summary of the Invention

[0004] Based on this, it is necessary for the present invention to provide a method and system for intelligent monitoring of soil erosion using UAV-assisted remote sensing technology, in order to solve at least one of the above-mentioned technical problems.

[0005] To achieve the above objectives, a method for intelligent monitoring of soil erosion using UAV-assisted remote sensing technology includes the following steps:

[0006] Step S1: Obtain geographic data of the target area and divide the area into monitoring grids based on digital elevation data and vegetation cover in the geographic data; determine the task allocation sequence of the UAV group based on the erosion sensitivity and connectivity of each monitoring grid.

[0007] Step S2: Control the UAV group to enter the corresponding monitoring grid according to the task assignment sequence in order to collect remote sensing image data;

[0008] Step S3: Stitch and register remote sensing image data collected by different UAVs to obtain multi-source remote sensing fusion data; classify and identify land surface types based on the multi-source remote sensing fusion data;

[0009] Step S4: Calculate the surface disturbance intensity parameters of each monitoring grid based on the changes in land surface type and texture feature variation in the same monitoring grid in adjacent image frames of the multi-source remote sensing fusion data.

[0010] Step S5: Combine the pre-acquired regional meteorological observation data to weight and correct the surface disturbance intensity parameters of each monitoring grid in order to determine the dynamic assessment results of soil and water loss for each monitoring grid.

[0011] This application constructs a monitoring grid with erosion sensitivity levels based on digital elevation data and vegetation cover, and reconstructs the grid by incorporating spatial connectivity. This makes the monitoring units more consistent with the spatial distribution characteristics of actual soil erosion. Furthermore, it introduces an adaptive task allocation mechanism for UAVs to achieve priority coverage of highly sensitive areas and multi-area collaborative inspection, effectively improving monitoring efficiency and resource utilization. Simultaneously, by performing stitching and geometric correction on remote sensing images acquired by multiple UAVs under a unified spatiotemporal reference, it eliminates scale differences and positional offsets between data from different platforms, improving the fusion efficiency of multi-source data. Consistency and accuracy; furthermore, by combining changes in surface type and texture features to construct the surface disturbance amplitude, and combining the temporal change rate of surface type proportion to form the disturbance rate, on this basis, multiple factors such as erosion sensitivity, slope aspect change and rainfall intensity are introduced for stratified weighted correction, so as to realize the dynamic characterization of soil erosion process from structural changes to evolution trend. Thus, under complex terrain and variable weather conditions, soil erosion monitoring with higher timeliness and precision can be achieved, the ability to identify hidden erosion and sudden erosion processes can be improved, and more reliable decision-making basis can be provided for regional ecological restoration and risk early warning.

[0012] Optionally, this application also provides an intelligent monitoring system for soil erosion using UAV-assisted remote sensing technology, for executing the intelligent monitoring method for soil erosion using UAV-assisted remote sensing technology as described above. The intelligent monitoring system for soil erosion using UAV-assisted remote sensing technology includes:

[0013] The task allocation module is used to acquire geographic data of the target area and divide the monitoring grid according to the digital elevation data and vegetation coverage in the geographic data; and determine the task allocation sequence of the UAV group according to the erosion sensitivity and connectivity of each monitoring grid.

[0014] The data acquisition module is used to control the UAV group to enter the corresponding monitoring grid according to the task assignment sequence in order to collect remote sensing image data;

[0015] The image stitching module is used to stitch and register remote sensing image data collected by different UAVs to obtain multi-source remote sensing fusion data; and to classify and identify land surface types based on the multi-source remote sensing fusion data.

[0016] The disturbance intensity calculation module is used to calculate the surface disturbance intensity parameters of each monitoring grid based on the changes in surface type and the magnitude of changes in texture features of the same monitoring grid in adjacent image frames of multi-source remote sensing fusion data.

[0017] The soil and water loss assessment module is used to combine pre-acquired regional meteorological observation data to weight and correct the surface disturbance intensity parameters of each monitoring grid in order to determine the dynamic assessment results of soil and water loss for each monitoring grid.

[0018] This application discloses an intelligent soil erosion monitoring system based on UAV collaborative remote sensing technology. This system can implement any of the UAV collaborative remote sensing technologies for intelligent soil erosion monitoring in this application. It serves as a medium for the operation and signal transmission between various modules to complete the intelligent soil erosion monitoring method based on UAV collaborative remote sensing technology. The modules within the system cooperate with each other, thereby enabling more timely detection of potential erosion areas in practical applications, improving the monitoring response speed and early warning accuracy, while reducing the ecological governance costs caused by monitoring delays. Attached Figure Description

[0019] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0020] Figure 1 This is a flowchart illustrating the steps of the intelligent monitoring method for soil erosion using UAV-assisted remote sensing technology of the present invention.

[0021] Figure 2 This is a schematic diagram of the feature correspondence points of adjacent image frames in an embodiment of the present invention;

[0022] Figure 3 This is a block diagram of the intelligent soil erosion monitoring system based on UAV collaborative remote sensing technology of the present invention;

[0023] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0024] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0025] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0026] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0027] To achieve the above objectives, please refer to Figures 1 to 3 This invention provides an intelligent monitoring method for soil erosion using UAV-assisted remote sensing technology, the method comprising the following steps:

[0028] Step S1: Obtain geographic data of the target area and divide the area into monitoring grids based on digital elevation data and vegetation cover in the geographic data; determine the task allocation sequence of the UAV group based on the erosion sensitivity and connectivity of each monitoring grid.

[0029] In one embodiment of this application, digital elevation data, land use data, and vegetation cover raster data of the target area are first retrieved from an existing geographic information platform. After unifying the three types of data into the same coordinate system, the grid is divided according to a regular pattern with an initial grid size of 30m×30m. Then, the average slope value (e.g., 5° as the low slope threshold and 15° as the high slope threshold) and vegetation cover rate (e.g., 0.3 as the low cover threshold) are calculated in each grid, and the erosion sensitivity is classified by combining the slope aspect change index (characterized by the slope aspect standard deviation of the pixels in the grid). Grids with consistent sensitivity and spatial continuity are merged to form a monitoring grid. Then, inspection path segments are generated according to the principle of prioritizing high-sensitivity grids and continuous adjacent grids. At the same time, the remaining battery power of each UAV (e.g., ≥40% is required to perform long-distance tasks) and maximum range (e.g., 10km) are read, and the path segments are matched according to range accessibility to obtain a task allocation sequence.

[0030] Step S2: Control the UAV group to enter the corresponding monitoring grid according to the task assignment sequence in order to collect remote sensing image data;

[0031] In a further embodiment, the drones are scheduled to take off sequentially according to the task sequence. Each drone maintains a fixed flight altitude (e.g., 120m) and a forward overlap rate (80% forward overlap and 60% lateral overlap) when entering the corresponding monitoring grid. It simultaneously acquires images through airborne visible light and multispectral sensors and records timestamps and GPS coordinates. For monitoring grids with large areas, a round-trip flight path is used to complete full coverage scanning. The flight speed is controlled (e.g., 8m / s) to ensure image clarity. At the same time, the acquired data is cached in real time and transmitted back to the ground processing terminal.

[0032] Step S3: Stitch and register remote sensing image data collected by different UAVs to obtain multi-source remote sensing fusion data; classify and identify land surface types based on the multi-source remote sensing fusion data;

[0033] In a further embodiment, after acquiring images from multiple UAVs, overlapping image frames are first filtered based on image boundaries and coordinate information. Significant texture points (such as pixels with gray-level gradients greater than a set threshold) are extracted within the overlapping area, and cross-image matching is performed. Affine transformation parameters are calculated based on the matching points, and geometric correction is performed on each image. Bilinear interpolation is used to complete pixel resampling. Subsequently, continuous images are generated by stitching them together frame by frame according to spatial location. On this basis, the band reflectance of each pixel is extracted, and vegetation indices (such as NDVI>0.4 indicating vegetation), bare soil indices, and water indices are calculated. Combined with texture change parameters (such as gray-level contrast changes within a window), land surface type classification is completed, and the dominant type is statistically analyzed according to the monitoring grid.

[0034] Step S4: Calculate the surface disturbance intensity parameters of each monitoring grid based on the changes in land surface type and texture feature variation in the same monitoring grid in adjacent image frames of the multi-source remote sensing fusion data.

[0035] In a further embodiment, between images acquired at adjacent times, the proportion of pixels with changes in surface type within the same monitoring grid is statistically analyzed, and the magnitude of changes in texture features is calculated. Pixel proportions with changes exceeding a set threshold (e.g., contrast change exceeding 15%) are then selected. The two are weighted (e.g., 0.6 and 0.4) to obtain the disturbance magnitude. The rate of change of each type of pixel proportion between two times (e.g., normalized in days) is further calculated as the disturbance rate, thereby forming the surface disturbance intensity parameter for each monitoring grid.

[0036] Step S5: Combine the pre-acquired regional meteorological observation data to weight and correct the surface disturbance intensity parameters of each monitoring grid in order to determine the dynamic assessment results of soil and water loss for each monitoring grid.

[0037] In a further embodiment, after obtaining the disturbance amplitude and disturbance rate, topographic constraints are first corrected based on the degree of erosion sensitivity. For example, the disturbance amplitude is multiplied by an amplification factor of 1.2 for highly sensitive grids, and a suppression factor of 0.8 is introduced for low-sensitive grids. Then, combined with rainfall data obtained from meteorological stations (e.g., hourly rainfall exceeding 20 mm is considered heavy rainfall), the corresponding grids are further processed, and the disturbance amplitude is multiplied by 1.1. At the same time, the disturbance rate weight is increased from 0.4 to 0.6 in the subsequent fusion. Finally, the corrected disturbance amplitude and disturbance rate are weighted and fused to obtain a comprehensive disturbance value, and high, medium, and low risk levels are divided according to preset thresholds (e.g., 0.7, 0.4) to form a dynamic assessment result of soil erosion.

[0038] Optionally, the geographic data includes digital elevation data, land use data, and vegetation cover; wherein step S1, dividing the monitoring grid, includes:

[0039] The target area is divided into several grid units according to the preset spatial resolution. The digital elevation data and vegetation cover corresponding to the cropping results are extracted according to the spatial coordinates of the grid units. The surface attributes of each grid unit are identified according to the land use data.

[0040] In this embodiment, the target area is unified to the same coordinate system (e.g., WGS84 to UTM projection), and then divided according to a fixed resolution, such as using 30m×30m as the basic grid size. Each grid is used to extract the elevation value (which can be the average value of the pixels within the grid) and vegetation cover (e.g., the coverage rate in the range of 0~1 obtained by NDVI inversion) from the DEM by using the coordinates of its upper left corner or center point. Land use data is usually a raster or vector layer. Here, each grid can be labeled by spatial overlay, such as marked as cultivated land, forest land or bare land, thus forming a grid data structure with elevation + vegetation + surface attributes (which can be understood as a three-field attribute table).

[0041] The erosion sensitivity of each grid cell is assessed based on digital elevation data and vegetation cover. Connectivity zones between adjacent grid cells with consistent erosion sensitivity and surface properties are identified and merged into a monitoring grid.

[0042] In a further embodiment, the slope of each grid cell is calculated (e.g., using a 3×3 window difference method to obtain the slope value) and combined with the vegetation coverage rate for a simple determination. For example, grids with a slope greater than 15° and a coverage rate less than 0.3 are directly classified as high-sensitivity, and the rest are divided into medium or low-sensitivity according to intervals. Next, connectivity is determined, generally using an 8-neighborhood method. As long as adjacent grids are consistent in sensitivity level and surface attributes, they are considered to be classified into the same block. For discrete blocks with too small an area (e.g., less than 3 grid cells), they can be merged into neighboring blocks. Finally, these continuous blocks are used as monitoring grids, which reduces fragmentation and facilitates subsequent UAVs to perform tasks by region.

[0043] Optionally, assessing the erosion sensitivity of each grid cell includes:

[0044] Extract the slope parameters and aspect values ​​of each grid cell from the digital elevation data, and determine the aspect change index based on the aspect value changes of different pixels within the same grid cell;

[0045] In this embodiment, when processing digital elevation data, a 3×3 window is first used to calculate the slope value within each grid cell. Specifically, the slope is calculated based on the differences between neighboring pixels in the elevation data, using a difference method (such as the Sobel algorithm). For example, within a 30m×30m grid, the unit of slope can be set to degrees (0° to 45°), thus obtaining a slope value for each grid cell. Next, the aspect change magnitude within each grid cell is calculated. Here, the aspect change magnitude is calculated by comparing the aspect values ​​of pixels within the grid (e.g., azimuth angles ranging from 0° to 360°), using the standard deviation of the aspect of different pixels within the grid as the calculation standard. The aspect change index quantifies the aspect change based on these standard deviations, setting a threshold (e.g., a change greater than 10° is considered a large aspect change).

[0046] The erosion sensitivity of grid cells with slope parameters higher than a preset slope threshold and vegetation coverage lower than a preset coverage threshold is determined to be a high-sensitivity cell.

[0047] Grid cells with slope parameters below the slope threshold and vegetation coverage above the coverage threshold are classified as low-sensitivity cells in terms of erosion sensitivity.

[0048] In a further embodiment, the erosion sensitivity of each grid cell is determined based on these slope and vegetation cover parameters. For example, if a grid cell has a slope exceeding 15° and a vegetation cover below 0.2 (where vegetation cover can be inferred from the NDVI value), it is classified as a "highly sensitive" area. Conversely, if the slope is less than 5° and the vegetation cover is greater than 0.7, it is classified as a "lowly sensitive" area.

[0049] The erosion sensitivity of grid cells with slope aspect change index higher than the preset slope aspect change threshold is enhanced and corrected, and the enhancement and correction results are constrained and corrected according to the surface attributes of each grid cell.

[0050] In a further embodiment, for the aspect change index, if the aspect change of a certain grid exceeds a set threshold (e.g., the aspect standard deviation is greater than 15°), the erosion sensitivity of that grid will be enhanced and corrected (e.g., if a grid is already a high-sensitivity unit, it will be adjusted to an extremely high-sensitivity unit level). In this case, the enhanced erosion sensitivity level will be adjusted according to the surface properties of the grid; for example, the erosion sensitivity level of forest or wetland grids will be reduced (if a grid is already a low-sensitivity unit, its erosion sensitivity level will be maintained), ensuring that the correction result reasonably reflects the interaction between topography and vegetation.

[0051] Optionally, determining the task assignment sequence for the UAV group in step S1 includes:

[0052] The priority of each monitoring grid is ranked according to its erosion sensitivity, and inspection path segments with different priorities are constructed based on the spatial adjacency of the monitoring grids.

[0053] In this embodiment, the monitoring grids are prioritized based on their erosion sensitivity. For example, based on the primary erosion sensitivity within each grid, grids consisting entirely of extremely sensitive units are assigned a higher priority (e.g., level 1), grids with over 90% being highly sensitive units are assigned level 2, and grids with over 90% being low-sensitive units are assigned level 3. Next, by calculating the spatial adjacency relationships between grids (e.g., identifying adjacent grids based on the 8-neighborhood rule), grids with the same priority are merged into inspection path segments. In this way, the inspection task for the entire area is divided into multiple path segments, ensuring that high-risk areas are addressed first.

[0054] Acquire the equipment status data and operational capability parameters of each drone in the drone group, and determine the range of tasks that each drone can perform based on the equipment status data and operational capability parameters;

[0055] Based on the executable mission range of each UAV, calculate the range energy consumption of each UAV to reach each inspection path segment, and correct the range energy consumption by combining the remaining power of each UAV in the equipment status data.

[0056] In a further embodiment, the device status data and operational capability parameters of the drones are obtained by connecting to the management cloud platform of the drone group. The mission range of each drone is determined based on its individual device status data (such as remaining battery power, flight capability, etc.) and operational capabilities (such as maximum payload, flight time, etc.). For example, a drone with 40% remaining battery power and a maximum flight time of 50 minutes can perform a mission with a flight radius of 15km. Based on this information, it is determined which inspection path segments the drone can reach, and the distance to each path segment and the energy consumption required to fly the entire path segment are calculated. Furthermore, for energy consumption during flight, the impact of hovering or speed changes is considered. If the drone needs to perform hovering operations or low-speed flight during the mission (e.g., taking pictures or performing environmental detection), a correction factor is added (e.g., considering an additional 20% energy consumption). After these corrections, tasks are rationally allocated based on the drone's actual flight capability, energy consumption, and remaining battery power to ensure that the drone maintains sufficient battery power during the mission.

[0057] Based on the corrected range and energy consumption of each UAV, the corresponding inspection path segments are allocated, and the execution order of the inspection path segments is optimized to obtain the task allocation sequence of the UAV group.

[0058] In a further embodiment, each inspection path segment is rationally allocated based on range energy consumption and remaining battery power. For example, if a path segment is far from a certain drone and has high energy consumption, it may be assigned to another drone with sufficient battery power. Then, through optimization algorithms (such as greedy algorithms), the execution order of each path segment is adjusted to ensure that the entire task is completed in the shortest possible time and to avoid drones running out of battery power, ultimately generating a task allocation sequence for the drone group.

[0059] Optionally, the stitching and registration of remote sensing image data acquired by different UAVs in step S3 includes:

[0060] The overlapping area between adjacent image frames is calculated based on the boundary range and spatial coordinates of the image frames in the remote sensing image data.

[0061] Within the overlapping area, salient points of surface texture are selected, and salient points of surface texture are matched between different image frames to generate feature correspondence points between different images.

[0062] In this embodiment, the outer boundary of each frame of image is first represented by its GPS coordinates as a rectangular range (generally using the latitude and longitude of the four corner points). Then, the image pairs with spatial overlap are found by calculating the intersection of the rectangles. Only when the overlapping area accounts for more than 20% of the area of ​​a single frame, they are included in the subsequent processing. The location points with obvious textures are extracted in the overlapping area. They can be filtered by gray-level gradient or corner response value (for example, only points with a response value greater than 0.01 are retained). Matching is performed between the two images to obtain a set of one-to-one feature correspondence point pairs. These point pairs are essentially the alignment reference between the two images.

[0063] The registration transformation parameters between image frames are calculated based on the spatial positional relationship of feature corresponding points in different images, and geometric correction processing is performed on each image frame according to the registration transformation parameters.

[0064] The geometrically corrected image frames are stitched together one by one according to the preset stitching order to obtain remote sensing image data.

[0065] In a further embodiment, after obtaining these corresponding points, the spatial transformation relationship between images can be calculated. An affine transformation (including translation, rotation, and scaling) is used to fit a set of transformation parameters using the least squares method. Then, the pixels in the original image are remapped to a unified coordinate position according to these parameters. The pixel values ​​at the new position are calculated by interpolation (such as bilinear interpolation to ensure grayscale continuity). After all images have undergone this geometric correction, they are stitched together frame by frame according to a pre-set stitching order (from left to right according to the flight path or coordinates). At the same time, brightness smoothing is performed in the seam area to finally form a continuous and complete remote sensing image.

[0066] Figure 2 This is a schematic diagram of the feature correspondence points of adjacent image frames in an embodiment of the present invention; for example... Figure 2 As shown, the area enclosed by the cool-colored lines is the overlapping area. Within this overlapping area, the yellow dot group and the red dot group are two different sets of feature points.

[0067] Of particular importance, the geometric correction process includes:

[0068] The spatial mapping relationship of each image frame is determined based on the registration transformation parameters, and the spatial position of each pixel in the image frame is repositioned based on the spatial mapping relationship to obtain the pixel target position distribution data.

[0069] In this embodiment, the previously calculated registration transformation parameters are first applied to each frame of image, which is equivalent to recalculating the position of each pixel. Specifically, the original pixel coordinates are mapped to a unified coordinate system through a transformation matrix. For example, the original position (100, 200) is mapped to a non-integer position such as (102.3, 198.7), thereby forming a pixel target position distribution table.

[0070] Based on the pixel location distribution data, pixel grayscale values ​​are resampled, and neighbor pixel interpolation is performed on target locations that do not correspond to the original pixel center location to obtain preliminary corrected image frame data.

[0071] In a further embodiment, a new image is generated based on these new locations in the pixel target location distribution table. Since most target locations are not integer pixel centers, resampling is required. Here, bilinear interpolation (calculating the weighted value of the surrounding 4 pixels) is used. Alternatively, cubic convolution interpolation can be used when high accuracy is required. For example, when the target point falls between four pixels, a new gray value is calculated according to the distance ratio to ensure that the image is continuous and does not have jagged edges, thus obtaining a preliminarily aligned image.

[0072] The boundary gaps in the initial correction of the image frame data are compensated, and the edges of the boundary gaps are smoothed to obtain the remote sensing image data.

[0073] In a further embodiment, edge problems are addressed because blank areas (such as holes with a value of 0) often appear at the boundaries after geometric transformation. In this case, compensation can be made by neighborhood expansion or nearest neighbor filling, such as filling the area with the average value of the surrounding effective pixels within a 3×3 window. At the same time, a simple smoothing process is performed on the stitching edges (such as a linear transition within a 5-pixel width) to avoid obvious seams, ultimately resulting in continuous and well-aligned remote sensing images.

[0074] Optionally, the classification and identification of land surface units in step S3 includes:

[0075] The spectral reflectance of each pixel in the remote sensing image data is extracted, and the vegetation index, bare soil index and water body index are determined according to the ratio between different spectral reflectances to obtain spectral reflectance characteristics.

[0076] In this embodiment, the multispectral data is first separated by band, such as red light, near-infrared, and short-wave infrared. Each pixel has a set of reflectance values ​​(usually normalized to 0~1). Then, the index is calculated by the ratio between the bands. For example, the vegetation index (such as NDVI, with a common threshold of 0.4 or higher is considered vegetation) is calculated using near-infrared and red light. The bare soil index is obtained by combining short-wave infrared and red light, and the water body index (such as water body if greater than 0.3) is obtained by combining green light and near-infrared light. In this way, each pixel has a set of spectral labels.

[0077] Calculate the texture change parameters of the same pixel in remote sensing image data within a preset time window to obtain texture features;

[0078] In a further embodiment, a time dimension is introduced, and two or three images are taken consecutively at the same location (e.g., with an interval of 7 days). Texture changes are calculated within a fixed window (e.g., 5×5 pixels), such as the contrast or uniformity change in the gray-level co-occurrence matrix. When the change exceeds a set threshold (e.g., 15%), the texture is considered to have changed significantly. This step mainly supplements areas where the spectral changes are not obvious but the structure has changed.

[0079] The categories of different pixels are determined based on spectral reflectance characteristics and texture characteristics;

[0080] Based on the monitoring grid boundaries, the pixel determination results are spatially aggregated, the area proportion of different types of pixels in each monitoring grid is statistically analyzed, and the land surface type of the monitoring grid is determined based on the category with the largest proportion.

[0081] In a further embodiment, the spectrum and texture are then used together to determine the category. For example, a high NDVI and stable texture are classified as vegetation, a high bare soil index and large texture variation are classified as bare land, and a high water index is directly classified as water. Finally, the pixel results are summarized according to the monitoring grid boundary. For example, the proportion of each type of pixel in a grid is counted (e.g., vegetation 60%, bare soil 30%, water 10%), and the largest proportion is taken as the dominant type of the grid. At the same time, the proportion of secondary types can be retained for subsequent analysis.

[0082] Of particular importance is the ability to classify different pixels based on spectral reflectance and texture features, including:

[0083] Pixels whose vegetation index in the spectral reflectance features is higher than the preset vegetation threshold are identified as vegetation-type pixels.

[0084] Pixels whose bare soil index parameter is higher than the preset bare soil threshold and whose texture change parameter is greater than the preset texture change threshold are identified as bare soil type pixels.

[0085] Pixels whose water index parameters are higher than the preset water threshold are classified as water-type pixels.

[0086] In this embodiment, during actual classification, the index value corresponding to each pixel is first organized into a simple data structure, such as a triplet (vegetation index, bare soil index, water index), and then each pixel is judged individually. For vegetation, the vegetation index is used as the threshold for judgment. For example, an NDVI greater than 0.4 is considered a vegetation pixel. This threshold can be slightly adjusted based on regional experience (for example, it can be reduced to 0.3 in arid areas). This can quickly screen out most areas with obvious vegetation coverage. The judgment of bare soil is more stringent because bare soil is easily confused with buildings or shadows. Therefore, in addition to the bare soil index (for example, greater than 0.25), a texture change condition is also superimposed. For example, within a preset time window, when the texture contrast change exceeds 15%, the corresponding pixel is judged as having undergone significant texture change, thereby identifying bare land areas with recent surface disturbance (such as tilling or erosion), so as to avoid misjudging areas that are only similar in spectral features but have not undergone actual disturbance as changed areas. Identifying water bodies is relatively simple. Water body indices (such as NDWI) are used as thresholds for judgment. For example, if the value is greater than 0.3, it is directly judged as a water body. At the same time, in order to avoid misjudgment, a restriction can be added, such as the vegetation index of the same pixel cannot be very high at the same time. Each pixel is classified into the corresponding category as long as it meets the conditions of a certain category, and finally a classification result map is formed.

[0087] Optionally, the surface disturbance intensity parameters include disturbance amplitude and disturbance rate; the calculation of the surface disturbance intensity parameters for each monitoring grid in step S4 includes:

[0088] Screen the proportion of pixels with significant texture changes within the monitoring grid whose texture feature changes exceed a preset threshold.

[0089] In this embodiment, the texture change amplitude is calculated pixel by pixel within each monitoring grid. For example, based on the grayscale contrast change value of two consecutive frames of images within a 5×5 window, a threshold is set (e.g., the change amplitude exceeds 15%) to filter out these pixels with significant changes. Then, the proportion of these pixels in the entire grid is calculated. For example, if there are 900 pixels in a grid, and 150 of them meet the condition, then the proportion of significant texture changes is 0.167. This step mainly reflects whether the surface detail structure has been disturbed.

[0090] The percentage of pixels showing changes in land surface type within the monitoring grid is statistically analyzed, and the percentage of pixels showing changes in land surface type and the percentage of pixels showing significant changes in texture are weighted according to a preset weight ratio to obtain the disturbance amplitude.

[0091] In a further embodiment, we examine the changes in land surface type and compare the classification results of the two periods within the same grid. We count the percentage of pixels whose type has changed (e.g., pixels that change from vegetation to bare soil), assuming the percentage is 0.25. Then, we merge this percentage with the previously calculated percentage of texture change according to weights, which can be set to 0.6 and 0.4. The perturbation amplitude is then 0.6 × 0.25 + 0.4 × 0.167, thus considering both explicit and implicit changes.

[0092] The temporal rate of change of the proportion of surface type pixels in each area within the statistical monitoring grid is used as the perturbation rate.

[0093] In a further embodiment, the rate of change of the proportion of various types of pixels over time is used as the perturbation rate. For example, if the vegetation proportion decreases from 0.7 to 0.5 over an interval of 10 days, the rate of change can be simply calculated as (0.7-0.5) / 10=0.02 / day. The perturbation rate of the grid can be calculated for each of the main types and then the average or maximum value can be taken.

[0094] Optionally, the weighted correction of the surface disturbance intensity parameters for each monitoring grid in step S5 includes:

[0095] The surface disturbance intensity parameters are corrected by terrain constraints based on the erosion sensitivity of the grid cells within the monitoring grid.

[0096] In this embodiment, the disturbance amplitude and disturbance rate of each monitoring grid are calculated beforehand, and then the terrain is corrected by combining the erosion sensitivity of the monitoring grid. For example, the sensitivity is simply divided into three levels: ultra-high, high, and low. For ultra-high sensitive grids and high sensitive grids, the disturbance amplitude can be multiplied by an amplification factor (e.g., 1.2 for ultra-high and 1.0 for high), and for low sensitive grids, it can be multiplied by a suppression factor (e.g., 0.8). This can reflect the areas of the terrain that are prone to erosion.

[0097] Acquire regional meteorological observation data and extract regional rainfall intensity; amplify the disturbance amplitude of the terrain constraint correction results corresponding to the monitoring grids where the regional rainfall intensity is higher than the preset rainfall intensity threshold, and increase the disturbance rate weight in the preset fusion weight.

[0098] In a further embodiment, meteorological data is introduced by reading hourly or daily rainfall from nearby weather stations. For example, an hourly rainfall exceeding 20 mm is used as a heavy rainfall threshold. When this threshold is reached within a certain time period of a monitoring grid, the disturbance amplitude corrected in the previous step is enhanced again, for example, by multiplying it by a coefficient of 1.1. At the same time, when calculating the final comprehensive disturbance, the weight of the disturbance rate is increased from the original 0.4 to 0.6. This is equivalent to emphasizing the influence of the rate of change in the case of heavy rainfall, and finally obtaining a disturbance result that takes into account both topography and rainfall effects.

[0099] Of particular importance is that terrain constraint correction includes:

[0100] If the erosion sensitivity of a grid cell within the monitoring grid is a high-sensitivity cell and the slope aspect change index is higher than the slope aspect change threshold, then the perturbation amplitude weight is increased in the preset fusion weight.

[0101] In this embodiment, an initial set of fusion weights is first set for the disturbance amplitude and disturbance rate, such as 0.6:0.4. If a monitoring grid is determined to be highly sensitive or ultra-highly sensitive, and the slope aspect change index is also relatively large (e.g., the standard deviation exceeds 20°), the weight of the disturbance amplitude will be increased, for example from 0.6 to 0.7 (for ultra-high sensitivity, it will be increased from 0.6 to 0.75), and the disturbance rate will be reduced to 0.3 accordingly. This is because such areas are more prone to structural damage and require greater emphasis on the degree of structural change.

[0102] If the erosion sensitivity of a grid cell within the monitoring grid is low, then the value of the disturbance amplitude is suppressed and corrected.

[0103] In another embodiment, if the monitoring grid as a whole consists of low-sensitivity cells, the weights are generally not changed. Instead, the value of the disturbance amplitude is directly compressed, for example, by multiplying it by a coefficient of 0.8, to avoid being misjudged as an increased risk due to small local changes. This step is more focused on reducing noise in the results.

[0104] If the aspect change index of any monitoring grid is higher than the aspect change threshold, the disturbance amplitude weight is increased in the corresponding fusion weight.

[0105] In another embodiment, the slope aspect change itself is very drastic, and even if it is not a highly sensitive area, it will trigger a weight adjustment on its own. For example, when the slope aspect change index exceeds the threshold (such as 20°), the disturbance amplitude weight is increased from 0.6 to 0.7. This can compensate for the impact of complex terrain on erosion. The whole process can be understood as a dynamic fine-tuning based on fixed weights and triggered by conditions.

[0106] Optionally, step S5 may determine the dynamic assessment results of soil erosion for each monitoring grid, including:

[0107] The disturbance amplitude is used as the evaluation index for changes in land surface structure, and the disturbance rate is used as the evaluation index for changes in land surface trend. The evaluation index for changes in land surface structure and the evaluation index for changes in land surface trend are fused according to the fusion weight to obtain the comprehensive disturbance characterization value of each monitoring grid unit.

[0108] In this embodiment, the disturbance amplitude and disturbance rate of each monitoring grid are treated as two input quantities. First, a set of fusion weights are assigned to them, such as 0.6 and 0.4 under normal circumstances (the former corresponds to structural changes, and the latter corresponds to the trend of change). Then, a weighted combination is directly performed, for example, the comprehensive disturbance value = 0.6 × disturbance amplitude + 0.4 × disturbance rate. This step is essentially to combine the magnitude of the change and the speed of the change into a unified index, which is convenient for unified judgment later. If the weights have been adjusted in advance (such as for heavy rainfall or highly sensitive areas), the adjusted weights are directly used in the calculation here.

[0109] Based on the preset grading threshold, the comprehensive disturbance characterization value is graded and determined, and each monitoring grid is divided into high loss risk area, medium loss risk area and low loss risk area, generating the initial soil and water loss assessment results.

[0110] In a further embodiment, after obtaining the comprehensive disturbance value, it can be classified according to the threshold. For example, a value above 0.7 can be set as high risk, 0.4 to 0.7 as medium risk, and below 0.4 as low risk. For example, if a certain grid is calculated to be 0.75, it can be directly classified into the high water loss risk area. In order to avoid misjudgment caused by fluctuations in the critical value, a simple buffer can also be added (for example, keeping the original level unchanged within ±0.05). Finally, each monitoring grid will correspond to a risk level, thus forming a more intuitive water and soil loss distribution result map.

[0111] Please see Figure 3 This application also provides a UAV-assisted remote sensing technology-based intelligent monitoring system 100 for soil erosion, used to execute the UAV-assisted remote sensing technology-based intelligent monitoring method for soil erosion as described above. The UAV-assisted remote sensing technology-based intelligent monitoring system for soil erosion includes:

[0112] The task allocation module 101 is used to acquire geographic data of the target area and divide the monitoring grid according to the digital elevation data and vegetation coverage in the geographic data; and determine the task allocation sequence of the UAV group according to the erosion sensitivity and connectivity of each monitoring grid.

[0113] The data acquisition module 102 is used to control the UAV group to enter the corresponding monitoring grid according to the task assignment sequence in order to collect remote sensing image data;

[0114] The image stitching module 103 is used to stitch and register remote sensing image data collected by different UAVs to obtain multi-source remote sensing fusion data; and to classify and identify land surface types based on the multi-source remote sensing fusion data.

[0115] The disturbance intensity calculation module 104 is used to calculate the surface disturbance intensity parameters of each monitoring grid based on the changes in surface type and the magnitude of changes in texture features of the same monitoring grid in adjacent image frames of multi-source remote sensing fusion data.

[0116] The soil and water loss assessment module 105 is used to combine pre-acquired regional meteorological observation data to weight and correct the surface disturbance intensity parameters of each monitoring grid in order to determine the dynamic assessment results of soil and water loss for each monitoring grid.

[0117] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0118] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for intelligent monitoring of soil erosion using UAV-assisted remote sensing technology, characterized in that, Includes the following steps: Step S1: Obtain geographic data of the target area and divide the area into monitoring grids based on digital elevation data and vegetation cover in the geographic data; determine the task allocation sequence of the UAV group based on the erosion sensitivity and connectivity of each monitoring grid. Step S2: Control the UAV group to enter the corresponding monitoring grid according to the task assignment sequence in order to collect remote sensing image data; Step S3: Stitch and register remote sensing image data collected by different UAVs to obtain multi-source remote sensing fusion data; classify and identify land surface types based on the multi-source remote sensing fusion data; Step S4: Based on the changes in land surface type and the magnitude of changes in texture features of the same monitoring grid in adjacent image frames of the multi-source remote sensing fusion data, calculate the land surface disturbance intensity parameters of each monitoring grid; the land surface disturbance intensity parameters include disturbance magnitude and disturbance rate. Step S4 involves calculating the surface disturbance intensity parameters for each monitoring grid, including: Screen the proportion of pixels with significant texture changes within the monitoring grid whose texture feature changes exceed a preset threshold. The percentage of pixels showing changes in land surface type within the monitoring grid is statistically analyzed, and the percentage of pixels showing changes in land surface type and the percentage of pixels showing significant changes in texture are weighted according to a preset weight ratio to obtain the disturbance amplitude. The temporal rate of change of the proportion of surface type pixels in each area within the statistical monitoring grid is used as the perturbation rate. Step S5: Combine the pre-acquired regional meteorological observation data to weight and correct the surface disturbance intensity parameters of each monitoring grid in order to determine the dynamic assessment results of soil and water loss for each monitoring grid.

2. The intelligent monitoring method for soil erosion using UAV-assisted remote sensing technology according to claim 1, characterized in that, Geographic data includes digital elevation data, land use data, and vegetation cover; Step S1, which involves dividing the monitoring grid, includes: The target area is divided into several grid units according to the preset spatial resolution. The digital elevation data and vegetation cover corresponding to the cropping results are extracted according to the spatial coordinates of the grid units. The surface attributes of each grid unit are identified according to the land use data. The erosion sensitivity of each grid cell is assessed based on digital elevation data and vegetation cover. Connectivity zones between adjacent grid cells with consistent erosion sensitivity and surface properties are identified and merged into a monitoring grid.

3. The intelligent monitoring method for soil erosion using UAV-assisted remote sensing technology according to claim 2, characterized in that, Assessing the erosion sensitivity of each grid cell includes: Extract the slope parameters and aspect values ​​of each grid cell from the digital elevation data, and determine the aspect change index based on the aspect value changes of different pixels within the same grid cell; The erosion sensitivity of grid cells with slope parameters higher than a preset slope threshold and vegetation coverage lower than a preset coverage threshold is determined to be a high-sensitivity cell. Grid cells with slope parameters below the slope threshold and vegetation coverage above the coverage threshold are classified as low-sensitivity cells in terms of erosion sensitivity. The erosion sensitivity of grid cells with slope aspect change index higher than the preset slope aspect change threshold is enhanced and corrected, and the enhancement correction results are constrained and corrected according to the surface attributes of each grid cell.

4. The intelligent monitoring method for soil erosion using UAV-assisted remote sensing technology according to claim 1, characterized in that, Step S1, which determines the task assignment sequence for the UAV group, includes: The priority of each monitoring grid is ranked according to its erosion sensitivity, and inspection path segments with different priorities are constructed by combining the spatial adjacency relationship of the monitoring grids. Acquire the equipment status data and operational capability parameters of each drone in the drone group, and determine the range of tasks that each drone can perform based on the equipment status data and operational capability parameters; Based on the executable mission range of each UAV, calculate the range energy consumption of each UAV to reach each inspection path segment, and correct the range energy consumption by combining the remaining power of each UAV in the equipment status data. Based on the corrected range and energy consumption of each UAV, the corresponding inspection path segments are allocated, and the execution order of the inspection path segments is optimized to obtain the task allocation sequence of the UAV group.

5. The intelligent monitoring method for soil erosion using UAV-assisted remote sensing technology according to claim 1, characterized in that, Step S3 involves stitching and registering remote sensing image data acquired by different UAVs, including: The overlapping area between adjacent image frames is calculated based on the boundary range and spatial coordinates of the image frames in the remote sensing image data. Within the overlapping area, salient points of surface texture are selected, and salient points of surface texture are matched between different image frames to generate feature correspondence points between different images. The registration transformation parameters between image frames are calculated based on the spatial positional relationship of feature corresponding points in different images, and geometric correction processing is performed on each image frame according to the registration transformation parameters. The geometrically corrected image frames are stitched together one by one according to the preset stitching order to obtain remote sensing image data.

6. The intelligent monitoring method for soil erosion using UAV-assisted remote sensing technology according to claim 1, characterized in that, Step S3, which involves classifying and identifying land surface units, includes: The spectral reflectance of each pixel in the remote sensing image data is extracted, and the vegetation index, bare soil index and water body index are determined according to the ratio between different spectral reflectances to obtain spectral reflectance characteristics. Calculate the texture change parameters of the same pixel in remote sensing image data within a preset time window to obtain texture features; The categories of different pixels are determined based on spectral reflectance characteristics and texture characteristics; Based on the monitoring grid boundaries, the pixel determination results are spatially aggregated, the area proportion of different categories of pixels in each monitoring grid is statistically analyzed, and the land surface type of the monitoring grid is determined based on the category with the largest proportion.

7. The intelligent monitoring method for soil erosion using UAV-assisted remote sensing technology according to claim 1, characterized in that, Step S5 involves weighting and correcting the surface disturbance intensity parameters for each monitoring grid, including: The surface disturbance intensity parameters are corrected by terrain constraints based on the erosion sensitivity of the grid cells within the monitoring grid. Acquire regional meteorological observation data and extract regional rainfall intensity; amplify the disturbance amplitude of the terrain constraint correction results corresponding to the monitoring grids where the regional rainfall intensity is higher than the preset rainfall intensity threshold, and increase the disturbance rate weight in the preset fusion weight.

8. The intelligent monitoring method for soil erosion using UAV-assisted remote sensing technology according to claim 1, characterized in that, Step S5 determines the dynamic assessment results of soil and water loss for each monitoring grid, including: The disturbance amplitude is used as the evaluation index for changes in land surface structure, and the disturbance rate is used as the evaluation index for changes in land surface trend. The evaluation index for changes in land surface structure and the evaluation index for changes in land surface trend are fused according to the fusion weight to obtain the comprehensive disturbance characterization value of each monitoring grid unit. Based on the preset grading threshold, the comprehensive disturbance characterization value is graded and determined, and each monitoring grid is divided into high loss risk area, medium loss risk area and low loss risk area, generating the initial soil and water loss assessment results.

9. A smart monitoring system for soil erosion using UAV-assisted remote sensing technology, characterized in that, For implementing the intelligent monitoring method for soil erosion using UAV-assisted remote sensing technology as described in claim 1, the intelligent monitoring system for soil erosion using UAV-assisted remote sensing technology comprises: The task allocation module is used to acquire geographic data of the target area and divide the monitoring grid according to the digital elevation data and vegetation coverage in the geographic data; and determine the task allocation sequence of the UAV group according to the erosion sensitivity and connectivity of each monitoring grid. The data acquisition module is used to control the UAV group to enter the corresponding monitoring grid according to the task assignment sequence in order to collect remote sensing image data; The image stitching module is used to stitch and register remote sensing image data collected by different UAVs to obtain multi-source remote sensing fusion data; and to classify and identify land surface types based on the multi-source remote sensing fusion data. The disturbance intensity calculation module is used to calculate the surface disturbance intensity parameters of each monitoring grid based on the changes in surface type and the magnitude of changes in texture features of the same monitoring grid in adjacent image frames of multi-source remote sensing fusion data. The soil and water loss assessment module is used to combine pre-acquired regional meteorological observation data to weight and correct the surface disturbance intensity parameters of each monitoring grid in order to determine the dynamic assessment results of soil and water loss for each monitoring grid.

Images