Forest region monitoring method and system based on unmanned aerial vehicle inspection

Abstract

The invention provides a forest area monitoring method and system based on unmanned aerial vehicle routing inspection, and the method comprises the steps: firstly obtaining a historical routing inspection data set containing a geographic position identifier and a topographic feature parameter in a target forest area, and generating an initial routing inspection route indicating a flight path and an image collection node according to the historical routing inspection data set; then, an unmanned aerial vehicle carrying a multispectral sensor is called to carry out dynamic routing inspection according to an air route, a real-time monitoring image set composed of vegetation coverage images of a plurality of monitoring areas under different timestamps is obtained, and then feature extraction and anomaly detection are carried out on the real-time monitoring image set; according to the method, an image anomaly feature set containing vegetation and surface structure anomaly indexes is determined, and finally, a forest region monitoring optimization strategy is generated based on the image anomaly feature set, so that the unmanned aerial vehicle inspection frequency and image acquisition node space distribution are adjusted, and more efficient and accurate monitoring of a forest region is realized.

Description

Technical Field

[0001] The present invention relates to the field of drone technology, and in particular to a forest area monitoring method and system based on drone inspection. Background Art

[0002] In today's society, forest monitoring is crucial for protecting forest resources, maintaining the ecological environment, and preventing forest disasters. Traditional forest monitoring methods rely primarily on manual inspections, which have many significant drawbacks. Manual inspections are limited in scope, making it difficult to cover large areas of forest. Areas with complex terrain and inaccessible transportation are particularly difficult to access, resulting in numerous blind spots in monitoring. Furthermore, manual inspections are inefficient, requiring significant manpower, material resources, and time. They also lack real-time monitoring capabilities and are difficult to detect abnormalities within the forest.

[0003] With the advancement of technology, some regions have begun using fixed monitoring equipment for forest monitoring. While these devices can, to a certain extent, compensate for the limitations of manual inspections in terms of coverage and real-time performance, they are fixed in position, with a limited monitoring range and no flexibility to adapt to the actual conditions of the forest. Furthermore, fixed monitoring equipment is susceptible to environmental influences such as inclement weather and obstruction by trees, which can reduce monitoring effectiveness. Furthermore, fixed monitoring equipment can only capture images and video information from a fixed perspective, making it difficult to fully and accurately reflect the growth status of forest vegetation and changes in surface structure.

[0004] Furthermore, some existing drone inspection methods also have drawbacks when applied to forest monitoring. Most methods rely solely on simple aerial photography, lacking effective utilization of historical forest data. Inspection route planning lacks specificity and scientific rationality, and fails to fully consider the topographical characteristics and actual monitoring needs of different forest areas. In terms of data processing, these methods only perform basic image acquisition, lacking in-depth feature extraction and anomaly detection. This makes it impossible to promptly and accurately detect abnormal vegetation conditions and surface structure within the forest. Furthermore, existing drone inspections are unable to dynamically adjust inspection strategies based on the actual conditions of the forest, making it difficult to achieve efficient and accurate monitoring of the forest. Summary of the Invention

[0005] In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides a forest area monitoring method based on drone inspection, the method comprising: Obtaining a historical inspection data set of a target forest area, wherein the historical inspection data set includes geographic location identifiers of multiple monitoring areas and corresponding terrain characteristic parameters; Generate initial inspection routes for multiple monitoring areas based on the historical inspection data set, wherein the initial inspection routes are used to indicate the flight paths and image acquisition nodes of the drone in the target forest area; Invoking a drone equipped with a multispectral sensor to perform a dynamic inspection operation according to the initial inspection route to obtain a set of real-time monitoring images of the target forest area, the set of real-time monitoring images including vegetation coverage images of multiple monitoring areas at different timestamps; Performing feature extraction and anomaly detection processing on the real-time monitoring image set to determine an image anomaly feature set of the target forest area, wherein the image anomaly feature set includes vegetation state anomaly indicators and surface structure anomaly indicators; A forest area monitoring optimization strategy is generated based on the image abnormal feature set.

[0006] On the other hand, an embodiment of the present invention also provides a forest area monitoring system based on drone inspection, including a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0007] Based on the above aspects, the embodiment of the present invention generates an initial inspection route by acquiring a set of historical inspection data of the target forest area, calls a drone equipped with a multispectral sensor to perform dynamic inspection operations to obtain a set of real-time monitoring images, performs feature extraction and anomaly detection processing on the drone to determine a set of image anomaly features, and then generates a forest area monitoring optimization strategy, thereby improving the accuracy and efficiency of forest area monitoring. Using historical inspection data to generate the initial inspection route can make the drone flight path and image acquisition nodes more consistent with the actual terrain of the forest area. The drone equipped with a multispectral sensor performs dynamic inspections and can obtain vegetation coverage images of multiple monitoring areas at different timestamps, enriching the monitoring data dimension. The processing of the real-time monitoring image set can accurately determine the set of image anomaly features, providing a key basis for discovering potential problems in the forest area. The forest area monitoring optimization strategy generated based on the set of image anomaly features can dynamically adjust the drone's inspection frequency and the spatial distribution of image acquisition nodes, realize the rational allocation of forest area monitoring resources, effectively improve the timeliness and pertinence of forest area monitoring, and better meet the needs of refined forest area management and real-time monitoring compared to traditional forest area monitoring methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 It is a schematic diagram of the execution flow of the forest area monitoring method based on drone inspection provided by an embodiment of the present invention.

[0009] Figure 2 Schematic diagram of exemplary hardware and software components of a forest area monitoring system based on drone inspections provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0010] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 This is a flow chart of a forest area monitoring method based on drone inspection provided by an embodiment of the present invention. The forest area monitoring method based on drone inspection is introduced in detail below.

[0011] Step S110: Acquire a historical inspection data set of the target forest area, wherein the historical inspection data set includes geographical location identifiers and corresponding terrain characteristic parameters of multiple monitoring areas.

[0012] In this embodiment, the historical inspection data set is collected from a wide range of sources, including past forest inspection records, geographic information systems (GIS), and related topographic mapping data. The geographic locations of multiple monitoring areas are represented using a common geographic coordinate system, such as longitude and latitude coordinates (x, y) to accurately determine the location of each monitoring area on the earth. The corresponding terrain characteristic parameters encompass a variety of terrain-related information, such as elevation data, slope, and aspect. Elevation data, represented by h, reflects the altitude of the monitoring area; slope indicates the inclination of the ground, represented by θ; and aspect indicates the direction of the slope, represented by α.

[0013] For example, a large forest area is divided into n monitoring zones, each with its own unique geographic location and terrain parameters. For the i-th monitoring zone (i = 1, 2, …, n), its geographic location is (xi, yi), and its terrain parameters include elevation hi, slope θi, and aspect αi. This constitutes the historical inspection data set.

[0014] Step S120: generating initial inspection routes for multiple monitoring areas based on the historical inspection data set, wherein the initial inspection routes are used to indicate the flight path and image acquisition nodes of the UAV in the target forest area.

[0015] In this embodiment, after obtaining a set of historical inspection data for the target forest area, it is necessary to generate initial inspection routes for multiple monitoring areas based on this historical inspection data. The generation of the initial inspection routes requires comprehensive consideration of multiple factors to ensure that the drone can complete the inspection mission efficiently and safely.

[0016] Step S121: extracting elevation gradient data and obstacle distribution data from the terrain feature parameters, and determining a flight altitude adjustment threshold of the UAV in the vertical direction based on the elevation gradient data.

[0017] In this embodiment, to generate a reasonable initial inspection route, we first need to extract elevation gradient data and obstacle distribution data from the terrain characteristic parameters. Elevation gradient data reflects the vertical variation of the terrain and can be obtained by calculating the ratio of the elevation difference between adjacent monitoring areas to the horizontal distance. Assuming the elevations of two adjacent monitoring areas i and i+1 are hi and hi+1, respectively, and the horizontal distance is d, the elevation gradient g can be expressed as g = (hi+1-hi) / d.

[0018] Obstacle distribution data describes the location and range of obstacles in the target forest area that may affect drone flight, such as trees and buildings. The distribution of obstacles can be determined by analyzing image data from historical inspection data and combining it with terrain data from the geographic information system.

[0019] Based on the extracted elevation gradient data, the vertical altitude adjustment threshold of the drone is determined. This process includes the following steps: Step S1211: performing piecewise linear fitting processing on the elevation gradient data to generate a terrain undulation trend curve of the target forest area, wherein the terrain undulation trend curve includes the position coordinates of a plurality of elevation mutation points.

[0020] In this embodiment, to better analyze the terrain's undulations, the elevation gradient data requires piecewise linear fitting. Piecewise linear fitting involves dividing the entire terrain area into several small segments, using a linear function within each segment to approximate the terrain variations. This method generates a terrain undulation trend curve for the target forest area. During the fitting process, points with significant elevation changes are discovered; these points are known as elevation mutation points. Let the coordinates of these elevation mutation points be (xj, hj), where xj represents the horizontal position and hj represents the corresponding elevation.

[0021] Step S1212: Calculate the vertical height difference and horizontal distance ratio between adjacent elevation mutation points, and determine the maximum allowable climbing angle of the UAV in the corresponding interval by combining the maximum thrust parameter of the UAV and the power efficiency model.

[0022] In this embodiment, after obtaining the topographic trend curve and the coordinates of the elevation change points, the ratio of the vertical height difference to the horizontal distance between adjacent elevation change points is calculated. Assuming the elevations of two adjacent elevation change points j and j+1 are hj and hj+1, respectively, and the horizontal distance is dj, then the ratio of the vertical height difference to the horizontal distance is rj = (hj+1-hj) / dj, which is the elevation gradient. To convert the elevation gradient into an angle, trigonometric functions are used, and the inverse tangent function is used for unit conversion, resulting in the corresponding angle θj = arctan(rj).

[0023] The maximum allowable climb angle for the drone within the corresponding range is determined by combining the drone's maximum thrust parameter, Tmax, with a power efficiency model. The power efficiency model describes the relationship between the drone's energy consumption and thrust output under different flight conditions. By analyzing this model and the maximum thrust parameter, the maximum allowable climb angle θmax, which allows the drone to safely fly on different slopes, can be calculated. For example, within a certain range, the maximum allowable climb angle θmax1 is calculated based on the power efficiency model and the maximum thrust parameter, combined with the converted angle θj.

[0024] Step S1213: Dynamically adjust the flight altitude adjustment threshold according to the maximum allowable climbing angle, so that the minimum safe altitude of the UAV from the ground during the climbing process is always greater than the preset terrain margin.

[0025] In this embodiment, after determining the maximum allowable climbing angle of the UAV in different intervals, the flight altitude adjustment threshold is dynamically adjusted according to these angles. The preset terrain margin is to ensure that the UAV maintains a certain safe distance from the ground during flight to avoid collision with obstacles. Let the preset terrain margin be δ and the flight altitude adjustment threshold be Δh. In each interval, the flight altitude adjustment threshold is adjusted according to the maximum allowable climbing angle and terrain conditions, so that the minimum safe height hmin of the UAV from the ground during the climbing process always satisfies hmin>δ. For example, in a certain interval, the flight altitude adjustment threshold is calculated as Δh1 based on the maximum allowable climbing angle and terrain. At this time, when the UAV flies in this interval, the flight altitude should be adjusted according to this threshold to ensure that the minimum safe height from the ground is greater than the preset terrain margin.

[0026] Step S1214: If it is detected that the deviation between the current flight altitude and the flight altitude adjustment threshold exceeds a second preset threshold, the hovering operation of the UAV is triggered and the obstacle avoidance path is replanned.

[0027] In this embodiment, during drone flight, the deviation between the current flight altitude and the altitude adjustment threshold is monitored in real time. Let the current flight altitude be hcurrent, the altitude adjustment threshold be Δh, and the second preset threshold be ε. When |hcurrent - Δh| > ε, the deviation between the current flight altitude and the adjustment threshold is too large, potentially compromising drone flight safety. At this point, the drone is triggered to hover, maintaining a stationary state in mid-air. Then, based on the current position and obstacle distribution, the obstacle avoidance path is replanned to ensure the drone can continue its flight safely.

[0028] Step S122: constructing a three-dimensional spatial obstacle avoidance model of multiple monitoring areas based on the obstacle distribution data, wherein the three-dimensional spatial obstacle avoidance model is used to mark the dynamic obstacle area in the UAV flight path.

[0029] In this embodiment, after obtaining obstacle distribution data, this data is used to construct a three-dimensional obstacle avoidance model for multiple monitoring areas. This model combines the three-dimensional spatial information of the target forest area with the location and range of obstacles to intuitively display the dynamic obstacle areas in the drone's flight path.

[0030] First, the three-dimensional space of the target forest area is gridded, dividing it into several small spatial units. Each spatial unit is represented by three-dimensional coordinates (x, y, z), where x and y represent horizontal position and z represents vertical height. Then, based on the obstacle distribution data, the spatial units where obstacles are located are marked in the three-dimensional spatial grid. These marked spatial units are the dynamic obstacle areas.

[0031] For example, within a monitoring area, obstacle distribution data identifies the locations and heights of trees and buildings. In the 3D obstacle avoidance model, the spatial cells containing these trees and buildings are marked as dynamic obstacle zones. When planning flight paths, drones can refer to this 3D obstacle avoidance model to avoid these dynamic obstacle zones, ensuring flight safety.

[0032] Step S123: generating path constraints for an initial inspection route based on the flight altitude adjustment threshold and the three-dimensional obstacle avoidance model, wherein the path constraints include a maximum climb angle limit and a minimum safety distance threshold.

[0033] In this embodiment, after determining the flight altitude adjustment threshold and building a three-dimensional obstacle avoidance model, the path constraints for the initial inspection route are generated based on this information. Path constraints are rules that restrict the flight path of the drone to ensure its safety and effectiveness during flight.

[0034] The maximum climb angle limit is set based on the previously determined maximum allowable climb angle for the drone in different ranges. When planning a flight path, the drone's climb angle must not exceed the maximum allowable climb angle to ensure sufficient power and stability. Let the maximum allowable climb angle be θmax, and the maximum climb angle limit in the path constraint is θ ≤ θmax.

[0035] The minimum safe distance threshold is designed to ensure a safe distance between the drone and obstacles. A minimum safe distance dmin is set based on the dynamic obstacle areas marked in the 3D obstacle avoidance model. When planning a flight path, the distance between the drone and the dynamic obstacle areas must always be greater than dmin to avoid collisions.

[0036] For example, within a monitoring area, the maximum allowable climb angle is θmax2, and the minimum safe distance threshold is dmin2, determined based on the flight altitude adjustment threshold and the three-dimensional spatial obstacle avoidance model. When generating the initial inspection route, the path constraints require that the drone's climb angle cannot exceed θmax2, and the distance from the dynamic obstacle area must be greater than dmin2.

[0037] Step S124: performing waypoint optimization processing on the path constraint conditions in combination with the geographical location identifiers of the multiple monitoring areas to generate an initial inspection route covering all monitoring areas, wherein the image acquisition node density of the initial inspection route is positively correlated with the vegetation coverage density of the monitoring area.

[0038] In this embodiment, after the path constraints of the initial inspection route are obtained, waypoint optimization processing is performed on these constraints in combination with the geographical location identifiers of multiple monitoring areas to generate an initial inspection route covering all monitoring areas.

[0039] Waypoint optimization involves selecting the appropriate waypoints for the drone, ensuring that the route efficiently covers all monitoring areas, while satisfying the path constraints. Heuristic algorithms, such as genetic algorithms and ant colony algorithms, can be used for waypoint optimization.

[0040] Furthermore, considering the vegetation density of the monitored area, the density of image acquisition nodes along the initial inspection route is set to be positively correlated with the vegetation density of the monitored area. Vegetation density can be determined by analyzing vegetation images from historical inspection data and is denoted by ρ. Let ρi be the vegetation density of monitored area i and ni be the image acquisition node density. Then, ni = k * ρi, where k is the scaling factor.

[0041] For example, within a target forest area, there are three monitoring areas, A, B, and C, with vegetation cover densities ρA, ρB, and ρC, respectively, where ρA>ρB>ρC. Through waypoint optimization and consideration of vegetation cover density, the generated initial inspection route has the highest density of image acquisition nodes in monitoring area A, followed by monitoring area B, and the lowest in monitoring area C. This ensures that more images are collected in areas with high vegetation cover density, enabling more accurate monitoring of the forest's vegetation conditions.

[0042] Step S130: calling a drone equipped with a multispectral sensor to perform a dynamic inspection operation according to the initial inspection route to obtain a real-time monitoring image set of the target forest area, wherein the real-time monitoring image set includes vegetation coverage images of multiple monitoring areas at different time stamps.

[0043] In this embodiment, after generating an initial inspection route, a drone equipped with a multispectral sensor is deployed to perform dynamic inspections along this route, acquiring a collection of real-time monitoring images of the target forest area. Multispectral sensors can simultaneously capture images in multiple wavelengths, including visible light and near-infrared, providing richer vegetation information.

[0044] Step S131: acquiring current wind speed data and light intensity data in real time during the flight of the UAV, and adjusting the flight speed and attitude stability parameters of the UAV based on the current wind speed data.

[0045] In this embodiment, during the flight of the drone, current wind speed data and light intensity data are obtained in real time. The wind speed data can be obtained by a wind speed sensor installed on the drone, represented by v; the light intensity data can be obtained by a light sensor, represented by I.

[0046] Adjust the flight speed and attitude stability parameters of the drone based on the current wind speed data. The specific steps are as follows: Step S1311: pre-establishing a correlation model between wind speed data and the aerodynamic drag of the UAV, wherein the correlation model is used to predict the energy consumption rate of the UAV under different wind speeds.

[0047] In this embodiment, to accurately adjust the drone's flight parameters based on wind speed, a correlation model between wind speed data and the drone's aerodynamic drag is pre-established. This correlation model describes the relationship between wind speed and the drone's aerodynamic drag and can be established using experimental data and theoretical analysis. Assuming the wind speed is v and the drone's aerodynamic drag is Fd, the correlation model can be expressed as Fd = f(v), where f(v) is a function of wind speed.

[0048] This correlation model can be used to predict the energy consumption rate of a drone under different wind speeds. The energy consumption rate is related to aerodynamic drag and flight speed. Let the energy consumption rate be P and the flight speed be v'. Considering that power is calculated as the product of force and velocity, with force in Newtons (N), velocity in meters per second (m / s), and power in watts (W), the energy consumption rate P = Fd * v'.

[0049] Step S1312: Dynamically adjust the ground speed control parameters of the UAV according to the energy consumption rate, so that the UAV maintains a preset airspeed when flying against the wind and increases propulsion power to offset the impact of wind speed, and reduces propulsion power when flying with the wind and maintains consistency between the ground speed and the route plan.

[0050] In this embodiment, after obtaining the energy consumption rate under different wind speeds, the ground speed control parameters of the drone are dynamically adjusted according to these rates. The preset airspeed is the expected flight speed of the drone in a windless state, represented by v0.

[0051] When a drone flies against a headwind, the wind speed v is opposite to the flight direction. To maintain the preset airspeed v0, propulsion power must be increased to offset the wind's influence. Let vheadwind be the ground speed control parameter for headwind flight, where vheadwind = v0 + v. Simultaneously, the energy consumption rate at this point is calculated using the correlation model, and propulsion power is adjusted to meet energy requirements.

[0052] When the drone is flying with a tailwind, the wind speed v is in the same direction as the flight. In this case, the propulsion power can be reduced while maintaining ground speed consistency with the planned route. Let vtailwind be the ground speed control parameter for tailwind flight, then vtailwind = v0 - v. Similarly, the propulsion power is adjusted based on the correlation model.

[0053] Step S1313: Based on the inertial measurement unit, the pitch angle and roll angle of the drone are monitored in real time. When the pitch angle or roll angle exceeds a third preset threshold, an attitude stabilization algorithm is activated to compensate for the attitude deviation caused by wind speed.

[0054] In this embodiment, the drone's pitch and roll angles are monitored in real time during flight using an inertial measurement unit (IMU). The pitch angle represents the rotation angle of the drone around its horizontal axis, denoted by φ; the roll angle represents the rotation angle of the drone around its vertical axis, denoted by ψ.

[0055] When the pitch or roll angle exceeds the third preset threshold, it indicates that the wind speed has caused the drone's attitude to shift, and the attitude stabilization algorithm must be activated to compensate for this shift. Let's assume the third preset thresholds are φ0 and ψ0, respectively. When |φ| > φ0 or |ψ| > ψ0, the attitude stabilization algorithm is activated. The attitude stabilization algorithm adjusts the drone's propeller speed to change its attitude and restore it to a stable state.

[0056] Step S1314: Feedback the adjusted flight speed and attitude stability parameters to the flight control system to control the image acquisition stability of the multispectral sensor in a dynamic environment.

[0057] In this embodiment, after adjusting the flight speed and attitude stability parameters of the drone, these adjusted parameters are fed back to the flight control system. The flight control system controls the flight of the drone based on these parameters to ensure the stability of image acquisition by the multispectral sensor in dynamic environments.

[0058] For example, in a windy environment, the drone collects real-time wind speed and light intensity data. Based on the wind speed data, it adjusts flight speed and attitude stabilization parameters. When the pitch or roll angle exceeds a third preset threshold, the attitude stabilization algorithm is activated. Finally, the adjusted parameters are fed back to the flight control system, enabling the multispectral sensor to capture stable images.

[0059] Step S132: Dynamically adjust the exposure time and sensitivity of the multispectral sensor according to the light intensity data so that the brightness balance of the collected vegetation coverage image meets a preset threshold.

[0060] In this embodiment, after acquiring light intensity data, the multispectral sensor's exposure time and sensitivity are dynamically adjusted based on this data. Exposure time refers to the length of time the sensor receives light, represented by t; sensitivity refers to the sensor's sensitivity to light, represented by ISO.

[0061] The preset threshold is to ensure that the brightness balance of the collected vegetation cover image meets the set standard. Let the preset threshold be L0, and by adjusting the exposure time and sensitivity, the brightness balance L of the collected image satisfies L≥L0.

[0062] To improve the robustness of subsequent image segmentation, in addition to adjusting exposure time and sensitivity, additional illumination invariance processing, such as color correction, is performed. Color correction eliminates the effects of illumination on image color, ensuring that images captured under different lighting conditions retain similar color characteristics. Color correction can be performed using methods such as color space conversion and histogram matching. For example, an image can be converted from RGB to another color space (such as HSV), the brightness component adjusted, and then converted back to RGB. These processes ensure that the quality of the captured vegetation cover images remains high under varying lighting conditions.

[0063] Step S133: When the UAV arrives at the image acquisition node, the multispectral sensor is triggered to perform multi-angle image acquisition on the target monitoring area to obtain a composite spectral image containing the visible light band and the near-infrared band.

[0064] In this embodiment, when the drone flies along the initial inspection route and arrives at the image acquisition node, it triggers the multispectral sensor to capture multi-angle images of the target monitoring area. Multi-angle image acquisition can obtain information about the target monitoring area from different angles, improving the image information content and accuracy.

[0065] Images collected by multispectral sensors contain information from both visible and near-infrared bands, forming a composite spectral image. The visible light band provides information on the color and texture of the target monitoring area, while the near-infrared band reflects information such as the health and moisture content of the vegetation.

[0066] For example, at an image acquisition node in a monitoring area, the drone triggers the multispectral sensor to collect images of the monitoring area from different angles (such as directly above, diagonally above, etc.) to obtain a composite spectral image containing visible light bands and near-infrared bands.

[0067] Step S134: Associatively store the composite spectral image with the corresponding geographical location identifier and acquisition timestamp to form a real-time monitoring image set.

[0068] In this embodiment, after obtaining the composite spectral images, these composite spectral images are associatively stored with the corresponding geographical location identifiers and acquisition timestamps. The geographical location identifier can accurately locate the acquisition position of the image, and the acquisition timestamp can record the acquisition time of the image.

[0069] These pieces of information are stored in the database to form a real-time monitoring image set. The real-time monitoring image set includes vegetation coverage images of multiple monitoring areas at different timestamps.

[0070] For example, for the composite spectral image of a certain monitoring area, it is associatively stored in the database with the geographical location identifier (x, y) of this monitoring area and the acquisition timestamp t. In this way, a complete real-time monitoring image set is formed.

[0071] Step S135: If it is detected that the remaining power of the drone is lower than the first preset threshold, estimate the power required for returning based on the distance between the current flight position and the return path. If the remaining power is sufficient for returning, re-plan the shortest return path and interrupt the uncompleted image acquisition nodes. If the remaining power is not sufficient for returning, preferentially execute the image acquisition nodes of the key monitoring areas and then force the drone to return.

[0072] In this embodiment, during the flight of the drone, the remaining power of the drone is monitored in real time. Let the remaining power be E and the first preset threshold be E0. When E < E0, it means that the remaining power of the drone is low and corresponding processing needs to be carried out.

[0073] First, estimate the power required for returning based on the distance between the current flight position and the return path. Considering the real-time wind speed during flight and the adjusted power model, integrate the dynamic ground speed control parameter in step S1312 into the power estimation. According to the real-time wind speed and the power efficiency model of the drone, calculate the energy consumption per unit distance of the drone flying at the current wind speed. Let the current flight position be (xcurrent, ycurrent) and the end position of the return path be (xreturn, yreturn), calculate the distance d between the current position and the return point. Combining with the energy consumption per unit distance of flying at the current wind speed, estimate the power Ereturn required for returning.

[0074] If E ≥ Ereturn, it means that the remaining battery power is sufficient for return home. At this time, the shortest return path is replanned and the unfinished image acquisition nodes are interrupted. A path planning algorithm is used to comprehensively consider factors such as terrain, obstacles, and power consumption to find the shortest path from the current position to the return point. After planning the shortest path, the drone stops the unfinished image acquisition task and returns along the newly planned path. If E is less than Ereturn, it means that the remaining battery power is insufficient to support a direct return home. At this time, the image acquisition nodes of the key monitoring areas are executed first. The key monitoring areas are pre-set based on factors such as the importance of the forest area and historical anomalies. The drone collects images of the key monitoring areas in order of priority and is forced to return home before the battery is about to run out. This ensures that as much monitoring data as possible from the key areas can be obtained under limited power conditions.

[0075] Step S140: performing feature extraction and anomaly detection processing on the real-time monitoring image set to determine an image anomaly feature set of the target forest area, wherein the image anomaly feature set includes vegetation state anomaly indicators and surface structure anomaly indicators.

[0076] In this embodiment, after obtaining a set of real-time monitoring images of the target forest area, it is necessary to perform feature extraction and anomaly detection processing on these images to determine a set of abnormal image features of the target forest area.

[0077] Step S141: performing a preprocessing operation on the vegetation coverage image, wherein the preprocessing operation includes noise suppression processing and image enhancement processing to improve the contrast between the vegetation area and the surface area in the vegetation coverage image.

[0078] In this embodiment, because the vegetation cover image in the real-time monitoring image set may be affected by various factors, such as sensor noise and uneven lighting, preprocessing operations are required. First, noise suppression processing is performed, and a filtering algorithm is used to remove noise from the image. Common filtering algorithms include mean filtering and median filtering. Mean filtering replaces the grayscale value of each pixel in the image with the average grayscale value of the pixels in its neighborhood; median filtering replaces the grayscale value of a pixel with the median grayscale value of the pixels in its neighborhood. Through these filtering algorithms, random noise in the image can be effectively reduced, making the image smoother.

[0079] Next, image enhancement is performed to improve the contrast between vegetation and ground surfaces in the vegetation cover image. Methods such as histogram equalization can be used to adjust the image's grayscale histogram to achieve a more uniform grayscale distribution, thereby enhancing image contrast. This allows vegetation and ground surfaces to appear more clearly in the image, facilitating subsequent feature extraction.

[0080] For example, in an image of vegetation cover, the boundary between the vegetation and the ground surface may be unclear due to the presence of noise. After noise suppression, the noise in the image is effectively removed, and the image becomes clearer. Furthermore, after image enhancement, the contrast between the vegetation area and the ground surface is significantly improved, and the outline of the vegetation is more clearly defined.

[0081] Step S142: extracting vegetation texture features and surface edge features from the preprocessed vegetation cover image. The vegetation texture features are used to characterize the density variation trend of the vegetation canopy, and the surface edge features are used to identify the boundary position between the surface exposed area and the vegetation covered area.

[0082] In this embodiment, after completing the preprocessing of the vegetation cover image, vegetation texture features and surface edge features are extracted from the processed image. For the extraction of vegetation texture features, methods such as grayscale co-occurrence matrix can be used. The grayscale co-occurrence matrix describes the spatial distribution relationship of grayscale values ​​in the image. By calculating various statistics of the grayscale co-occurrence matrix, such as contrast, correlation, energy, etc., the characteristic values ​​of vegetation texture can be obtained. These characteristic values ​​can reflect the density change trend of the vegetation canopy. For example, a texture with higher contrast may indicate a higher density of the vegetation canopy.

[0083] Edge detection algorithms, such as the Canny edge detection algorithm, can be used to extract surface edge features. The Canny edge detection algorithm accurately detects edge information in vegetation-covered images by performing Gaussian smoothing, gradient calculation, non-maximum suppression, and double-threshold processing. In vegetation-covered images, this edge information can identify the boundary between exposed and covered areas.

[0084] For example, in a preprocessed vegetation cover image, vegetation texture features were extracted using a gray-level co-occurrence matrix (GLCM). High texture contrast in certain areas may indicate a higher density of vegetation canopy. Furthermore, the Canny edge detection algorithm was used to extract surface edge features, clearly identifying the boundaries between exposed and vegetation-covered areas.

[0085] Step S143: performing a difference comparison between the vegetation texture features and the baseline texture features of the corresponding monitoring area in the historical inspection data set to generate vegetation status abnormality indicators, which include the canopy density decrease amplitude and the proportion of leaf discoloration areas.

[0086] In this embodiment, after the vegetation texture features are extracted, they need to be compared with the baseline texture features of the corresponding monitoring area in the historical inspection data set to generate vegetation status abnormality indicators.

[0087] Step S1431: extracting the baseline texture features of the same monitoring area in the historical time period from the historical inspection data set, wherein the baseline texture features include a vegetation canopy grayscale distribution histogram and texture direction gradient statistics.

[0088] In this embodiment, baseline texture features for the same area as the currently monitored area over a historical time period are identified from a historical inspection data set. These baseline texture features include a vegetation canopy grayscale distribution histogram and texture directional gradient statistics. The vegetation canopy grayscale distribution histogram describes the distribution of grayscale values ​​within the vegetation canopy, reflecting the overall brightness characteristics of the vegetation canopy; the texture directional gradient statistics describe how the texture varies in different directions, reflecting the textural structural characteristics of the vegetation canopy.

[0089] For example, for monitoring area A, the grayscale distribution histogram and texture direction gradient statistics of the vegetation canopy in a certain period of time in the past are extracted from the historical inspection data set as the benchmark texture features of the monitoring area.

[0090] Step S1432: normalize the grayscale distribution histograms of the current vegetation texture feature and the benchmark texture feature, calculate the Bhattacharyya distance as the first difference index, and vectorize the texture direction gradient statistics and calculate the cosine similarity as the second difference index.

[0091] In this embodiment, to accurately compare the current vegetation texture features with the baseline texture features, their grayscale distribution histograms are first normalized. Normalization makes the grayscale distribution histograms of different images comparable, unifying the value ranges of the grayscale distribution histograms to the same interval. The Bhattacharyya distance between the normalized grayscale distribution histograms is then calculated. This distance measures the similarity between two probability distributions and serves as the first difference indicator.

[0092] For the texture directional gradient statistics, we first vectorize them, converting them into vector form. Then, we calculate the cosine similarity between the vectorized texture directional gradient statistics. Cosine similarity measures the cosine value of the angle between two vectors, reflecting their directional similarity, and serves as the second difference indicator.

[0093] For example, for the current vegetation texture features and the benchmark texture features in the monitoring area A, the Bhattacharyya distance of the grayscale distribution histogram calculated after normalization is d1, and the cosine similarity of the texture direction gradient statistics after vectorization is s1.

[0094] Step S1433: normalize the first difference index and the second difference index to the same dimension interval and perform weighted summation to generate a canopy density decrease amplitude.

[0095] In this embodiment, because the first difference metric (Bhattacharyya distance, range [0, 1]) and the second difference metric (cosine similarity, range [-1, 1]) have different dimensions, the second difference metric needs to be processed to combine them. First, the cosine similarity is corrected to the range [0, 1]. This can be done by taking its absolute value. The corrected cosine similarity is |s|, where s is the original cosine similarity.

[0096] The corrected first and second difference indices were then normalized using a linear normalization method, mapping their value ranges to the interval [0, 1]. Different weights were then assigned to the first and second difference indices based on their impact on the canopy density decrease. The weight of the first difference indices was set to w1, and the weight of the second difference indices to w2, with w1 + w2 = 1. The normalized first and second difference indices were weighted and summed to obtain the canopy density decrease.

[0097] For example, for the monitoring area A, after normalizing the Bhattacharyya distance d1 and the corrected cosine similarity |s1| to the interval [0, 1], the normalized d1' and |s1'| are obtained respectively. Assuming w1=0.6 and w2=0.4, the decrease in canopy density is 0.6*d1'+0.4*|s1'|.

[0098] Step S1434: Identify the leaf area in the current vegetation coverage image through an image segmentation algorithm, extract the color space distribution characteristics of the leaf area and compare them with the reference color characteristics to determine the proportion of the leaf discoloration area.

[0099] In this embodiment, in order to determine the proportion of leaf discoloration areas, an image segmentation algorithm is first used to identify the leaf surface areas in the current vegetation coverage image. The image segmentation algorithm can separate different areas in the image and separate the leaf surface areas from the background. Then, the color space distribution characteristics of the leaf surface area are extracted, such as the mean and variance in the RGB color space. The extracted color space distribution characteristics are compared with the baseline color characteristics of the corresponding monitoring area in the historical inspection data set to determine which areas have changed color. The proportion of leaf surface discoloration areas is determined by counting the ratio of the area of ​​the discolored area to the total area of ​​the leaf surface area.

[0100] For example, in the current vegetation cover image of monitoring area A, an image segmentation algorithm is used to segment the leaf surface area and extract its color spatial distribution characteristics. After comparing it with the baseline color characteristics, it is found that the color of some leaf areas has changed. The area of ​​these discolored areas is calculated as the proportion of the total leaf area to obtain the percentage of leaf discoloration areas.

[0101] Step S1435: If the canopy density decrease or the proportion of leaf discoloration areas exceeds a fourth preset threshold, the corresponding monitoring area is marked as an abnormal vegetation status area.

[0102] In this embodiment, a fourth preset threshold is set to determine whether the vegetation status in the monitoring area is abnormal. When the canopy density decreases or the proportion of leaf discoloration exceeds this threshold, it indicates that the vegetation in the monitoring area may be abnormal and is marked as an abnormal vegetation status area.

[0103] For example, for monitoring area A, if the calculated decrease in canopy density or the proportion of leaf discoloration areas exceeds the fourth preset threshold, monitoring area A will be marked as an area with abnormal vegetation status.

[0104] Step S144: detecting surface structure anomaly indicators based on the surface edge features, wherein the surface structure anomaly indicators include an estimated value of the surface collapse depth and the number of newly added illegal paths.

[0105] In this embodiment, after extracting surface edge features, surface structural anomaly indicators are detected based on these features. To estimate the depth of a surface subsidence, stereoscopic image information or multi-view image analysis can be utilized. By comparing the position and shape changes of the surface edge in the current image with those in historical images, combined with topographic data, the depth of the surface subsidence can be estimated.

[0106] To detect the number of newly added illegal paths, image recognition technology can be used to identify newly added paths in the current image that are not normally planned, and the number of these paths is counted as the number of newly added illegal paths.

[0107] For example, in images of monitoring area B, by comparing the current image with historical images, significant changes in the surface edge were detected. Combined with topographic data, the depth of the surface subsidence was estimated. Image recognition technology was also used to identify newly added illegal paths in the image and count their number.

[0108] Step S145: performing a spatiotemporal correlation analysis on the vegetation state abnormality index and the surface structure abnormality index to determine a set of image abnormality features of the target forest area.

[0109] In this embodiment, after obtaining the vegetation anomaly indicator and the surface structure anomaly indicator, they need to be subjected to spatiotemporal correlation analysis. Spatiotemporal correlation analysis considers the temporal and spatial relationships between these anomaly indicators. The vegetation anomaly indicator and the surface structure anomaly indicator can be matched based on the geographic location of the monitoring area and the time of acquisition. The analysis is performed to determine whether vegetation anomalies and surface structure anomalies occur simultaneously at the same time and location, and the degree of correlation between them.

[0110] For example, in a certain monitoring area C, at the same time, vegetation status anomaly indicators showed a significant decrease in canopy density, while surface structure anomaly indicators indicated surface subsidence. Through spatiotemporal correlation analysis, it was determined that this monitoring area had a relatively serious anomaly and was included in the image anomaly feature set for the target forest area.

[0111] Step S150: generating a forest area monitoring optimization strategy based on the image abnormality feature set, wherein the forest area monitoring optimization strategy is used to adjust the inspection frequency of the UAV and the spatial distribution of the image acquisition nodes.

[0112] In this embodiment, after determining the abnormal image feature set of the target forest area, it is necessary to generate a forest area monitoring optimization strategy based on these abnormal features to improve the efficiency and accuracy of forest area monitoring.

[0113] Step S151: determining the fire risk level of the target monitoring area according to the canopy density decrease in the vegetation status abnormality index, and evaluating the probability of pest and disease spread based on the proportion of the leaf discoloration area.

[0114] In this embodiment, the fire risk level of the target monitoring area is determined based on the degree of decrease in canopy density, a component of the vegetation abnormality indicator. A greater decrease in canopy density indicates poorer vegetation health, a higher likelihood of vegetation becoming dry and flammable, and a greater fire risk. The degree of canopy density decrease can be divided into different ranges, each corresponding to a fire risk level.

[0115] The probability of pest spread is assessed based on the percentage of leaf discoloration. A larger percentage indicates more severe damage from pests and diseases, and a higher likelihood of pest spread. A mapping can be established to map the percentage of leaf discoloration to a range of pest spread probability.

[0116] For example, for monitoring area D, if the decrease in canopy density is in a certain higher range, the fire risk level of the monitoring area will be determined as high risk; if the proportion of leaf discoloration areas is large, the probability of pest and disease spread in the monitoring area will be assessed as high based on the mapping relationship.

[0117] Step S152: generating a soil erosion warning signal based on the estimated value of the surface subsidence depth in the surface structure anomaly indicator, and identifying the activity level of illegal logging based on the number of new illegal paths.

[0118] In this embodiment, a soil erosion warning signal is generated based on the estimated surface subsidence depth in the surface structure anomaly indicator. The greater the surface subsidence depth, the more severe the soil erosion. Different surface subsidence depth thresholds can be set. When the estimated value exceeds a certain threshold, a soil erosion warning signal of the corresponding level is generated.

[0119] The level of illegal logging activity is determined based on the number of new illegal routes. The greater the number of new illegal routes, the more likely illegal logging activity is. The level of illegal logging activity can be categorized into different levels based on the number of new illegal routes.

[0120] For example, in monitoring area E, if the estimated depth of surface collapse exceeds a certain higher threshold, a high-level soil erosion warning signal is generated; if the number of new illegal paths is large, the activity level of illegal logging in the monitoring area is identified as high.

[0121] Step S153: normalize the fire risk level, pest and disease spread probability, soil erosion warning signal, and illegal logging activity level into sub-risk scores within a unified scoring interval, and perform weighted summation of the sub-risk scores according to a preset weight coefficient to generate a comprehensive risk score for each monitoring area.

[0122] In this example, to comprehensively assess the risk of each monitoring area, it is necessary to normalize the fire risk level, pest and disease spread probability, soil erosion warning signal, and illegal logging activity into sub-risk scores within a unified scoring range. Linear normalization can be used to map these different risk indicators to the same scoring range, such as [0, 100].

[0123] The sub-risk scores are then weighted and summed according to preset weighting factors. These weighting factors are pre-determined based on the importance of these risk factors to forest safety. For example, the fire risk level is weighted as w3, the probability of pest spread as w4, the soil erosion warning signal as w5, and the level of illegal logging as w6, where w3 + w4 + w5 + w6 = 1. The normalized sub-risk scores are multiplied by their corresponding weighting factors and then summed to obtain a comprehensive risk score for each monitored area.

[0124] For example, for monitoring area F, the fire risk level, probability of pest and disease spread, soil erosion warning signal and illegal logging activity are normalized to obtain sub-risk scores s2, s3, s4 and s5 respectively. Assuming w3=0.3, w4=0.2, w5=0.3 and w6=0.2, the comprehensive risk score of the monitoring area is 0.3*s2+0.2*s3+0.3*s4+0.2*s5.

[0125] Step S154: Adjusting the drone inspection frequency of the corresponding monitoring area based on the comprehensive risk score, so that the inspection frequency of the monitoring area with a larger comprehensive risk score is higher than the inspection frequency of the monitoring area with a smaller comprehensive risk score.

[0126] In this example, to more effectively monitor the forest area, the frequency of drone inspections in the corresponding monitoring area is adjusted based on the comprehensive risk score. A higher comprehensive risk score indicates a higher likelihood of abnormal conditions in the monitoring area, and more frequent inspections are required.

[0127] Step S1541: allocating an initial inspection frequency to each monitoring area, wherein the initial inspection frequency is associated with the geographical area and vegetation type of the monitoring area.

[0128] In this embodiment, before adjusting the inspection frequency, an initial inspection frequency is assigned to each monitoring area. This initial inspection frequency is related to the geographic area and vegetation type of the monitoring area. Generally speaking, monitoring areas with larger geographic areas may require more inspections to cover the entire area. Different vegetation types have different sensitivities to environmental changes, so monitoring areas with vegetation types that are susceptible to pests, diseases, or fires may require a higher initial inspection frequency.

[0129] For example, for monitoring area G, which has a large geographical area and a vegetation type of flammable coniferous forest, a relatively high initial inspection frequency is assigned to it; while for monitoring area H, which has a small geographical area and a vegetation type of relatively stable broad-leaved forest, a relatively low initial inspection frequency is assigned to it.

[0130] Step S1542: establishing a mapping relationship table between the comprehensive risk score and the inspection frequency, wherein the mapping relationship table is used to indicate the inspection frequency with exponential growth corresponding to the comprehensive risk score.

[0131] In this embodiment, in order to accurately adjust the inspection frequency based on the comprehensive risk score, a mapping relationship table between the comprehensive risk score and the inspection frequency is established. This mapping relationship table reflects the exponential growth relationship between the comprehensive risk score and the inspection frequency, that is, the higher the comprehensive risk score, the faster the inspection frequency increases.

[0132] For example, when the comprehensive risk score is in a lower range, the inspection frequency increases less; when the comprehensive risk score is in a higher range, the inspection frequency increases more. This exponential growth mapping relationship can more rationally allocate inspection resources.

[0133] Step S1543: When the comprehensive risk score of the monitoring area exceeds the fifth preset threshold for multiple consecutive inspection cycles, an emergency inspection mode is triggered and the inspection frequency of the monitoring area is adjusted according to the mapping relationship table.

[0134] In this embodiment, in order to promptly detect and handle abnormal situations in high-risk areas, the fifth preset threshold is set to a dynamic value. By calculating the mean and standard deviation of the historical comprehensive risk score, the fifth preset threshold is set to the mean of the historical score plus the standard deviation. Assume that the set of historical comprehensive risk scores is {score1, score2, ..., scoren}, then the historical score mean_score is the sum of all scores divided by the number of scores, that is, mean_score=(score1+score2+...+scoren) / n, and the standard deviation std_score can be obtained by calculating the square root of the average of the sum of the squares of the differences between each score and the mean. The fifth preset threshold, threshold, is equal to mean_score+std_score.

[0135] If the comprehensive risk score of a monitored area exceeds this dynamic fifth preset threshold for multiple consecutive inspection cycles, it indicates that the monitored area has a serious abnormal risk, triggering emergency inspection mode. In emergency inspection mode, the inspection frequency of the monitored area is quickly adjusted according to the mapping table, and the number of inspections is increased.

[0136] For example, for monitoring area I, if its comprehensive risk score exceeds the dynamic fifth preset threshold for three consecutive inspection cycles, the emergency inspection mode is triggered and its inspection frequency is greatly increased according to the mapping relationship table.

[0137] Step S1544: Synchronize the adjusted inspection frequency to the UAV's task scheduling system, and prioritize the execution of monitoring area nodes whose comprehensive risk score is ranked above a preset percentile in the task queue.

[0138] In this embodiment, after adjusting the inspection frequency of a monitoring area, the adjusted inspection frequency is synchronized with the drone's task scheduling system. The task scheduling system schedules drone inspection tasks based on the new inspection frequency. Furthermore, within the task queue, priority is given to monitoring area nodes whose comprehensive risk scores rank above a preset percentile. This preset percentile is set based on actual needs; for example, monitoring area nodes with comprehensive risk scores in the top 20% are prioritized.

[0139] For example, in the task queue, the monitoring area nodes ranked in the top 20% of comprehensive risk scores are placed in front, and drone inspections are prioritized to ensure that high-risk areas can be monitored in a timely manner.

[0140] Step S155: Dynamically adjust the density of image acquisition nodes in the monitoring area according to the spatial distribution data of the image acquisition nodes and the comprehensive risk score.

[0141] In this example, to more accurately capture anomaly information in forest areas, the density of image acquisition nodes in the monitored area is dynamically adjusted based on their spatial distribution data and the overall risk score. This allows drones to collect more images in areas with higher overall risk, improving the accuracy of anomaly detection.

[0142] For example, step S1551: mapping the comprehensive risk score to a risk level coefficient within a preset range.

[0143] In this embodiment, to facilitate subsequent calculations and adjustments, the comprehensive risk score needs to be mapped to a risk level coefficient within a preset interval. First, the preset interval is determined, for example, set to [0, 1]. Then, a mapping relationship between the comprehensive risk score and the risk level coefficient is established. A linear mapping method can be used. Assuming the comprehensive risk score ranges from [min_score to max_score], for a monitoring area with a comprehensive risk score of current_score, its corresponding risk level coefficient, risk_coefficient, can be calculated as follows: First, the relative position of the comprehensive risk score within the range is calculated as (current_score - min_score) / (max_score - min_score). The resulting value is the corresponding risk level coefficient for the monitoring area. In this way, monitoring areas with different comprehensive risk scores are mapped to different risk level coefficients within the preset interval. For example, if monitoring area A has a high comprehensive risk score, the resulting risk level coefficient after mapping is close to 1; whereas, monitoring area B has a low comprehensive risk score, its risk level coefficient is close to 0.

[0144] Step S1552: extract the spatial distribution data of the current image acquisition nodes, and calculate the initial value of the node density of each monitoring area. The initial value of the node density is the number of image acquisition nodes per unit area.

[0145] In this embodiment, to adjust the density of image acquisition nodes, it is first necessary to extract the spatial distribution data of the current image acquisition nodes. These data record the specific location of each image acquisition node in the monitoring area. Then calculate the initial value of the node density of each monitoring area, which is measured by the number of image acquisition nodes per unit area. For each monitoring area, first determine its area area, and then count the number of image acquisition nodes node_count in the area, then the initial value of the node density initial_density is equal to node_count divided by area. For example, the area of ​​monitoring area C is large, and the number of image acquisition nodes contained in it is relatively small, so its initial value of node density is low; while the area of ​​monitoring area D is small, but the number of image acquisition nodes is large, so its initial value of node density is high.

[0146] Step S1553: normalize the risk level coefficient and the initial value of node density to a preset ratio range respectively, and generate a density adjustment factor according to the ratio of the risk level coefficient to the initial value of node density. The density adjustment factor is positively correlated with the risk level coefficient and negatively correlated with the initial value of node density.

[0147] In this embodiment, in order to ensure that the risk level coefficient and the initial value of node density can be calculated at the same scale, they need to be normalized to a preset ratio interval, such as [0, 1]. For the risk level coefficient, since it is already within the preset interval [0, 1], it does not need to be normalized again. For the initial value of node density, assuming that its value range is [min_density, max_density], for a certain monitoring area, the initial value of node density is current_initial_density, then the normalized initial value of node density normalized_initial_density can be calculated by (current_initial_density-min_density) / (max_density-min_density).

[0148] To avoid the problem of the density adjustment factor being infinite due to the initial node density being normalized to 0, a minimum density threshold ε (e.g., ε = 1e-5) is set. When the initial node density after normalization is less than ε, its value is set to ε.

[0149] Next, a density adjustment factor is generated based on the ratio of the risk level coefficient to the initial normalized node density value. Let risk_coefficient be the risk level coefficient and normalized_initial_density' be the initial normalized node density value. Then, density_adjustment_factor equals risk_coefficient divided by normalized_initial_density'. Since a higher risk level coefficient indicates a greater risk in the monitoring area, requiring an increase in image acquisition node density, while a higher initial node density value indicates that the area's node density is already high and may not require a significant increase, the density adjustment factor is positively correlated with the risk level coefficient and negatively correlated with the initial node density value. For example, if monitoring area E has a higher risk level coefficient but a lower initial node density value, the calculated density adjustment factor will be larger, indicating a significant increase in image acquisition node density. On the other hand, monitoring area F has a lower risk level coefficient and a higher initial node density value, resulting in a smaller density adjustment factor and likely requiring only a small or no increase in node density.

[0150] Step S1554: If the newly added image acquisition nodes generated by the density adjustment factor overlap with the obstacle area, the newly added image acquisition nodes are allocated in the adjacent area using a spatial interpolation method until the non-conflict condition is met.

[0151] In this embodiment, after determining the image acquisition nodes to add based on the density adjustment factor, it is necessary to check whether the locations of these newly added nodes overlap with obstacle areas. Obstacle areas are marked in the previously constructed three-dimensional obstacle avoidance model and include areas such as trees and buildings that may affect drone flight and image acquisition. If the newly added image acquisition nodes are found to overlap with the obstacle area, spatial interpolation is used to allocate the new image acquisition nodes to the adjacent area.

[0152] Spatial interpolation is a method for estimating unknown point data based on known point data. In this scenario, the adjacent area is selected, centered around the newly added image acquisition node that overlaps with the obstacle. Based on the spatial location of the adjacent area and the distribution of image acquisition nodes, an appropriate location is estimated to assign the new node. For example, linear interpolation is used to calculate a new location in the adjacent area that does not overlap with the obstacle, based on the coordinates and related attributes of the adjacent known nodes. This process is repeated until all newly added image acquisition nodes meet the no-conflict condition, meaning they do not overlap with the obstacle area.

[0153] Step S1555: determining the density increment of the image acquisition nodes in each monitoring area based on the density adjustment factor, marking the monitoring areas with node density increment greater than 0 as priority adjustment areas, and generating the spatial coordinates of the newly added image acquisition nodes.

[0154] In this embodiment, after obtaining the density adjustment factor, the image acquisition node density increment for each monitoring area is determined based on it. The node density increment is calculated by multiplying the density adjustment factor by a preset coefficient: increment = density_adjustment_factor * coefficient. This preset coefficient is arbitrarily set based on actual conditions and experience to control the magnitude of the node density increase.

[0155] Monitoring areas with node density increments greater than 0 are marked as priority adjustment areas. This is because these areas require an increase in image acquisition node density to better monitor forest conditions. Then, based on the node density increment and the spatial distribution of the monitoring area, the spatial coordinates of the newly added image acquisition nodes are generated.

[0156] When generating the spatial coordinates of the newly added image acquisition nodes, a path planning algorithm (such as the A algorithm) is used to ensure the feasibility of the route, taking into account the drone's minimum turning radius and endurance. The A algorithm is a heuristic search algorithm that finds the optimal path by evaluating the cost of each node (including the actual cost from the starting point to the current node and the estimated cost from the current node to the target node). In this scenario, the location of the newly added image acquisition node is used as the target node, and considering constraints such as the drone's minimum turning radius and endurance, the A* algorithm is used to search for feasible paths and node locations. For example, when generating the coordinates of the newly added nodes, it is ensured that the distance between adjacent nodes meets the drone's minimum turning radius requirements, and at the same time, it is ensured that the drone has sufficient power to return after completing the image acquisition tasks for all nodes. This can avoid the problem of theoretically generated node coordinates being unattainable in actual flight.

[0157] Step S1556: Modify the initial inspection route according to the spatial coordinates of the newly added image acquisition nodes, so that the density of the adjusted image acquisition nodes is nonlinearly positively correlated with the comprehensive risk score of the priority adjustment area.

[0158] In this embodiment, after the spatial coordinates of the newly added image acquisition nodes are generated, the initial inspection route needs to be revised based on these coordinates. The initial inspection route was previously generated based on historical inspection data and factors such as terrain, but due to the addition of new image acquisition nodes, the initial inspection route needs to be adjusted.

[0159] The goal of this adjustment is to ensure that the density of the adjusted image acquisition nodes is nonlinearly positively correlated with the comprehensive risk score of the priority adjustment area. This means that the higher the comprehensive risk score, the greater the increase in the density of image acquisition nodes, but it is not a simple linear relationship. For example, nonlinear functions such as exponential or logarithmic functions can be used to describe this relationship. When correcting the route, the location of the newly added image acquisition nodes and the flight performance of the drone are taken into consideration, and the flight path of the drone is replanned so that the drone can collect images according to the adjusted image acquisition node density. For example, for priority adjustment areas with higher comprehensive risk scores, more stopovers are added to the route to allow the drone to collect more images in these areas.

[0160] Step S1557: If it is detected that the spatial coordinates of the newly added image acquisition node overlap with the dynamic obstacle area in the obstacle distribution data, the density increments of the image acquisition nodes in the adjacent monitoring areas are reallocated and the density adjustment factor is updated.

[0161] In this embodiment, after revising the initial inspection route, it is necessary to recheck whether the spatial coordinates of the newly added image acquisition nodes overlap with the dynamic obstacle areas in the obstacle distribution data. Dynamic obstacle areas are obstacles that may change over time, such as moving vehicles and animals. If overlap is detected, the image acquisition node density increments in adjacent monitoring areas need to be reallocated.

[0162] Specifically, the density increments corresponding to newly added image acquisition nodes that overlap with dynamic obstacle areas are distributed to adjacent monitoring areas according to pre-set rules. For example, the distribution can be based on factors such as the risk level and area of ​​the adjacent monitoring areas. After the distribution is completed, the density adjustment factors of the adjacent monitoring areas are updated. The ratio of the risk level coefficient of these monitoring areas to the normalized initial node density is recalculated to obtain a new density adjustment factor. The image acquisition node density increments and the spatial coordinates of the newly added nodes are then re-determined based on the new density adjustment factor until all newly added image acquisition nodes do not overlap with dynamic obstacle areas, completing the dynamic adjustment process of the entire image acquisition node density.

[0163] In summary, this embodiment can effectively monitor forest areas, detect abnormal situations in a timely manner, and adjust the inspection frequency and image acquisition node density according to risk conditions, thereby improving the efficiency and accuracy of forest area monitoring.

[0164] Figure 2 A schematic diagram illustrates exemplary hardware and software components of a drone-based forest monitoring system 100 that can implement the concepts of the present application, as provided in some embodiments of the present application. For example, a processor 120 can be used in the drone-based forest monitoring system 100 to perform the functions described in the present application.

[0165] The forest monitoring system 100 based on drone inspection can be a general-purpose server or a special-purpose server, both of which can be used to implement the forest monitoring method based on drone inspection in this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0166] For example, the forest area monitoring system 100 based on drone inspection may include a network port 110 connected to the network, one or more processors 120 for executing program instructions, a communication bus 130, and different forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the forest area monitoring system 100 based on drone inspection may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present application can be implemented according to these program instructions. The forest area monitoring system 100 based on drone inspection also includes an I / O interface 150 between the computer and other input and output devices.

[0167] For ease of explanation, only one processor is described in the forest area monitoring system 100 based on drone inspection. However, it should be noted that the forest area monitoring system 100 based on drone inspection in the present application may also include multiple processors, so the steps performed by one processor described in the present application may also be performed jointly or individually by multiple processors. For example, if the processor of the forest area monitoring system 100 based on drone inspection executes step A and step B, it should be understood that step A and step B may also be performed jointly by two different processors or individually in one processor. For example, the first processor executes step A, the second processor executes step B, or the first processor and the second processor execute steps A and B together.

[0168] In addition, an embodiment of the present invention also provides a readable storage medium, in which computer-executable instructions are preset. When the processor executes the computer-executable instructions, the above-mentioned forest area monitoring method based on drone inspection is implemented.

[0169] It should be noted that in order to simplify the description of the present invention and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, multiple features are sometimes combined into one embodiment, figure or description thereof.

Claims

1. A forest area monitoring method based on drone inspection, characterized in that: The method comprises: Obtaining a historical inspection data set of a target forest area, wherein the historical inspection data set includes geographic location identifiers of multiple monitoring areas and corresponding terrain characteristic parameters; Generate initial inspection routes for multiple monitoring areas based on the historical inspection data set, wherein the initial inspection routes are used to indicate the flight paths and image acquisition nodes of the drone in the target forest area; Invoking a drone equipped with a multispectral sensor to perform a dynamic inspection operation according to the initial inspection route to obtain a set of real-time monitoring images of the target forest area, the set of real-time monitoring images including vegetation coverage images of multiple monitoring areas at different timestamps; Performing feature extraction and anomaly detection processing on the real-time monitoring image set to determine an image anomaly feature set of the target forest area, wherein the image anomaly feature set includes vegetation state anomaly indicators and surface structure anomaly indicators; A forest area monitoring optimization strategy is generated based on the image abnormal feature set.

2. The forest area monitoring method based on drone inspection according to claim 1 is characterized in that: Generating initial inspection routes for multiple monitoring areas based on the historical inspection data set includes: Extracting elevation gradient data and obstacle distribution data from the terrain feature parameters, and determining a flight altitude adjustment threshold of the UAV in a vertical direction based on the elevation gradient data; Constructing a three-dimensional spatial obstacle avoidance model of multiple monitoring areas based on the obstacle distribution data, wherein the three-dimensional spatial obstacle avoidance model is used to mark dynamic obstacle areas in the flight path of the UAV; Generate path constraints for an initial inspection route based on the flight altitude adjustment threshold and the three-dimensional obstacle avoidance model, wherein the path constraints include a maximum climb angle limit and a minimum safety distance threshold; The path constraint conditions are optimized with respect to the geographical location identifiers of the multiple monitoring areas to generate an initial inspection route covering all monitoring areas. The image acquisition node density of the initial inspection route is positively correlated with the vegetation coverage density of the monitoring area.

3. The forest area monitoring method based on drone inspection according to claim 2 is characterized in that: The step of determining a flight altitude adjustment threshold of the UAV in a vertical direction based on the elevation gradient data includes: Performing piecewise linear fitting processing on the elevation gradient data to generate a terrain undulation trend curve of the target forest area, wherein the terrain undulation trend curve includes position coordinates of a plurality of elevation mutation points; Calculate the vertical height difference and horizontal distance ratio between adjacent elevation mutation points, and determine the maximum allowable climb angle of the UAV in the corresponding interval by combining the UAV's maximum thrust parameters and power efficiency model; Dynamically adjusting the flight altitude adjustment threshold according to the maximum allowable climb angle so that the minimum safe altitude of the UAV from the ground during the climb is always greater than a preset terrain margin; If it is detected that the deviation between the current flight altitude and the flight altitude adjustment threshold exceeds a second preset threshold, the hovering operation of the drone is triggered and the obstacle avoidance path is replanned.

4. The forest area monitoring method based on drone inspection according to claim 1 is characterized in that: The calling of the drone equipped with the multispectral sensor to perform a dynamic inspection operation according to the initial inspection route to obtain a set of real-time monitoring images of the target forest area includes: Acquire current wind speed data and light intensity data in real time during the flight of the drone, and adjust the flight speed and attitude stability parameters of the drone based on the current wind speed data; Dynamically adjusting the exposure time and sensitivity of the multispectral sensor according to the light intensity data so that the brightness balance of the collected vegetation cover image meets a preset threshold; When the UAV arrives at the image acquisition node, it triggers the multispectral sensor to collect multi-angle images of the target monitoring area and obtain a composite spectral image containing visible light bands and near-infrared bands; The composite spectral image is associated with the corresponding geographic location identifier and acquisition timestamp and stored to form a real-time monitoring image set; If it is detected that the remaining battery power of the drone is lower than the first preset threshold, the power required for return is estimated based on the distance between the current flight position and the return path. If the remaining battery power is sufficient for return, the shortest return path is replanned and the unfinished image acquisition nodes are interrupted. If the remaining battery power is insufficient for return, the image acquisition nodes of the key monitoring areas are executed first and then forced to return.

5. The forest area monitoring method based on drone inspection according to claim 4 is characterized in that: The adjusting the flight speed and attitude stability parameters of the UAV based on the current wind speed data includes: Pre-establishing a correlation model between wind speed data and the aerodynamic drag of the UAV, wherein the correlation model is used to predict the energy consumption rate of the UAV under different wind speeds; Dynamically adjusting the ground speed control parameters of the drone based on the energy consumption rate, so that the drone maintains a preset airspeed and increases propulsion power to offset the wind speed when flying against a headwind, and reduces propulsion power and maintains ground speed consistent with the planned route when flying with a tailwind; monitoring the pitch angle and roll angle of the drone in real time based on an inertial measurement unit, and activating an attitude stabilization algorithm to compensate for attitude deviation caused by wind speed when the pitch angle or roll angle exceeds a third preset threshold; The adjusted flight speed and attitude stability parameters are fed back to the flight control system to control the image acquisition stability of the multispectral sensor in a dynamic environment.

6. The forest area monitoring method based on drone inspection according to claim 1 is characterized in that: The performing feature extraction and anomaly detection processing on the real-time monitoring image set to determine an abnormal feature set of the image of the target forest area includes: performing a preprocessing operation on the vegetation cover image, wherein the preprocessing operation includes noise suppression processing and image enhancement processing to improve the contrast between the vegetation area and the surface area in the vegetation cover image; Extracting vegetation texture features and surface edge features from the preprocessed vegetation cover image, wherein the vegetation texture features are used to characterize the density change trend of the vegetation canopy, and the surface edge features are used to identify the boundary position between the surface exposed area and the vegetation covered area; Comparing the vegetation texture features with the baseline texture features of the corresponding monitoring area in the historical inspection data set to generate vegetation status abnormality indicators, which include the canopy density decrease and the proportion of leaf discoloration areas; Detecting surface structure anomaly indicators based on the surface edge features, wherein the surface structure anomaly indicators include an estimated value of surface collapse depth and a number of newly added illegal paths; A spatiotemporal correlation analysis is performed on the vegetation state abnormality index and the surface structure abnormality index to determine a set of image abnormality features of the target forest area.

7. The forest area monitoring method based on drone inspection according to claim 6 is characterized in that: The step of comparing the vegetation texture features with the baseline texture features of the corresponding monitoring area in the historical inspection data set to generate vegetation status abnormality indicators includes: Extracting the baseline texture features of the same monitoring area in the historical time period from the historical inspection data set, the baseline texture features including the vegetation canopy grayscale distribution histogram and texture direction gradient statistics; Normalizing the grayscale distribution histograms of the current vegetation texture feature and the benchmark texture feature, calculating the Bhattacharyya distance as a first difference indicator, and vectorizing the texture direction gradient statistics and calculating the cosine similarity as a second difference indicator; Normalizing the first difference index and the second difference index to the same dimension interval and then performing weighted summation to generate a canopy density decrease amplitude; The image segmentation algorithm is used to identify the leaf area in the current vegetation cover image, extract the color space distribution characteristics of the leaf area and compare them with the reference color characteristics to determine the proportion of the leaf discoloration area; If the canopy density decreases or the proportion of leaf discoloration areas exceeds a fourth preset threshold, the corresponding monitoring area is marked as an abnormal vegetation status area.

8. The forest area monitoring method based on drone inspection according to claim 6 is characterized in that: The generating of the forest area monitoring optimization strategy based on the image abnormal feature set includes: Determine the fire risk level of the target monitoring area based on the canopy density decrease in the vegetation abnormality indicator, and assess the probability of pest spread based on the proportion of leaf discoloration areas; generating a soil erosion warning signal based on the estimated depth of surface subsidence in the surface structure anomaly indicator, and identifying the activity level of illegal logging based on the number of new illegal paths; Normalize the fire risk level, pest and disease spread probability, soil erosion warning signal, and illegal logging activity level into sub-risk scores within a unified scoring range, and perform a weighted summation of the sub-risk scores based on a preset weight coefficient to generate a comprehensive risk score for each monitoring area; Adjusting the drone inspection frequency of the corresponding monitoring area based on the comprehensive risk score, so that the inspection frequency of the monitoring area with a higher comprehensive risk score is higher than the inspection frequency of the monitoring area with a lower comprehensive risk score; The density of image acquisition nodes in the monitoring area is dynamically adjusted according to the spatial distribution data of the image acquisition nodes and the comprehensive risk score.

9. The forest area monitoring method based on drone inspection according to claim 8 is characterized in that: The adjusting the drone inspection frequency of the corresponding monitoring area based on the comprehensive risk score includes: Assigning an initial inspection frequency to each monitoring area, wherein the initial inspection frequency is associated with the geographical area and vegetation type of the monitoring area; Establishing a mapping relationship table between the comprehensive risk score and the inspection frequency, wherein the mapping relationship table is used to indicate the inspection frequency with exponential growth corresponding to the comprehensive risk score; When the comprehensive risk score of the monitoring area exceeds a fifth preset threshold for multiple consecutive inspection cycles, an emergency inspection mode is triggered and the inspection frequency of the monitoring area is adjusted according to the mapping relationship table; The adjusted inspection frequency is synchronized to the UAV's task scheduling system, and monitoring area nodes with comprehensive risk scores ranked above the preset percentile are prioritized in the task queue.

10. A forest monitoring system based on drone inspection, characterized in that: It includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the forest area monitoring method based on drone inspection as described in any one of claims 1 to 9.

Images