Forestry resource dynamic monitoring system based on unmanned aerial vehicle patrol

Abstract

The invention is applied to the technical field of forestry monitoring, and discloses a forestry resource dynamic monitoring system based on unmanned aerial vehicle patrol, and the system comprises an unmanned aerial vehicle patrol cluster module which comprises a master control unmanned aerial vehicle and a slave control unmanned aerial vehicle, and the master control unmanned aerial vehicle and the slave control unmanned aerial vehicle are interconnected through a wireless communication network; the unmanned aerial vehicle carries a multispectral imaging sensor, a laser radar (LiDAR), an infrared thermal imager and a positioning navigation module, and is used for collecting real-time image data, topographic data and environmental parameters of forestry resources. According to the forestry resource dynamic monitoring system based on unmanned aerial vehicle patrol, through a dynamic task allocation strategy and a real-time abnormal response mechanism, when the electric quantity of the unmanned aerial vehicle is insufficient or data is interrupted, the system automatically triggers task reallocation and eliminates patrol blind areas, and compared with traditional single-sortie operation or preset route operation, the patrol coverage rate is improved from lower than 90% to 98% or above, and the unmanned aerial vehicle patrol efficiency is improved. The data overlapping rate is reduced from higher than 40% to lower than 15%, and resource waste and monitoring omission risks are remarkably reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of forestry monitoring, in particular to a forestry resource dynamic monitoring system based on unmanned aerial vehicle patrol. BACKGROUND

[0002] As the core carrier of the terrestrial ecosystem, forestry resources bear key functions such as ecological protection, biodiversity conservation, and carbon sink regulation. The dynamic monitoring accuracy and response efficiency of forestry resources are directly related to the implementation of national ecological safety and sustainable development strategy of forestry. With the intensification of global climate change and the expansion of human activities, risks such as forest fires, illegal logging, and pest spread frequently occur, which puts forward the stereoscopic demand of "real-time sensing, accurate identification, intelligent early warning, and efficient disposal" for forestry resource monitoring. However, the existing forestry resource monitoring technology system still has multi-dimensional bottlenecks and cannot adapt to the requirements of intelligent forestry management under the new situation. Traditional forestry monitoring has long relied on a combination of manual patrol and ground station observation. This mode has significant shortcomings: manual patrol is limited by complex terrain (such as mountainous areas and marshes) and weather conditions. The daily patrol range is usually less than 500 hectares, and the coverage period of large areas of forest is several months. Data recording relies on manual judgment, which is highly subjective and has a high error rate. Ground stations can obtain accurate local environmental parameters, but the station layout density is limited, and they cannot reflect the spatial heterogeneity characteristics of the forest area. Moreover, data updates lag behind the dynamic process of vegetation growth and disaster evolution. Such methods can only achieve "point-line" level monitoring and cannot meet the needs of global and dynamic monitoring. Although the application of satellite remote sensing technology has broken through the spatial coverage limit, it faces the dual contradiction of resolution and timeliness: optical satellite image resolution is usually above 10 meters, which cannot identify single-tree-level disease or small-scale illegal logging behavior. Moreover, it is affected by cloud cover and the effective data acquisition rate in rainy forest areas is less than 40%. Synthetic aperture radar (SAR) can operate all-weather, but data interpretation is difficult, and the accuracy of identifying vegetation health is usually less than 75%. More importantly, the satellite revisit period is usually 3-15 days, which cannot capture real-time events such as early-stage fires and pest outbreaks, often missing the best disposal opportunity. In recent years, unmanned aerial vehicle technology has been gradually applied to forestry monitoring due to its strong flexibility and high resolution. However, the existing solutions have not formed a systematic capability and mainly have the following defects: Cluster coordination and inefficient task scheduling: Current UAV patrols are mostly single sortie operations or simple formation flights, lacking intelligent coordination mechanisms. Although UAV clusters achieve wide coverage, they do not use dynamic task allocation strategies. When some UAVs run out of power or data transmission is interrupted, blind areas in the patrol are likely to occur. The UAV honeycomb system supports multi-vehicle coordination, but the path planning relies on pre-set routes and cannot dynamically adjust the patrol priority based on real-time abnormal data (such as sudden fire points). The emergency response time exceeds 30 minutes, making it difficult to meet the needs of early disaster disposal. Existing technologies generally struggle to balance the contradiction between patrol coverage (usually less than 90%) and data overlap rate (often higher than 40%), leading to resource waste or missed monitoring. Insufficient precision of multi-source data fusion: The data generated by multi-spectral sensors, LiDAR, infrared thermal imagers, and other devices carried by UAVs have temporal and spatial heterogeneity. However, existing fusion technologies lack standardized processing procedures, and time synchronization often relies on the UAV's own clock, with errors often exceeding 100ms, making it difficult to accurately match spectral data and point cloud data at the same time. Spatial registration often uses pixel-level matching algorithms, with registration errors in complex terrain areas reaching several meters, directly affecting subsequent analysis accuracy. Additionally, data repair methods for low-density point clouds (<30 points / square meter) and noisy images are limited, with simple interpolation methods often used, resulting in a confidence level of less than 60% in completing data, making it difficult to support high-precision monitoring needs. Weak dynamic analysis and intelligent early warning capabilities: Existing systems rely on traditional image processing algorithms for change detection, such as threshold segmentation methods, which have an accuracy rate of only 70%-85% for identifying changes in forest coverage, and cannot distinguish between natural wilting and human-caused deforestation. Tree species classification often uses machine learning methods such as support vector machines (SVM), with an accuracy rate of less than 85% in mixed forest areas, far from meeting the needs of fine-grained management. Early warning mechanisms are often triggered by a single indicator (such as temperature detection for fires), without integrating multi-dimensional information such as spectral features and terrain parameters, resulting in a false alarm rate of over 25%. Additionally, after an early warning, there is a lack of closed-loop linkage with patrol equipment, requiring manual re-planning of re-patrol tasks, with a significant response lag. As the coverage of digital twin forest is included in the assessment indicators of forestry carbon sink capacity, the shortcomings of existing technology systems are becoming increasingly prominent. There is an urgent need to break through the technical bottlenecks in the "collection-fusion-analysis-early warning-scheduling" chain and develop a dynamic monitoring system that integrates cluster coordination patrol, high-precision data fusion, deep learning analysis, and intelligent early warning and decision-making. This will enable a transition from "passive response monitoring" to "active predictive control" of forestry resources, providing technical support for forest disaster prevention and control, ecological restoration, and carbon sink accounting. SUMMARY

[0003] The present application aims to provide a forestry resource dynamic monitoring system based on unmanned aerial vehicle patrol, to solve the problems of limited coverage of artificial patrol, strong subjectivity of data recording, insufficient density of ground station layout and inability to reflect spatial heterogeneity of forest area, data update lag, contradiction between resolution and timeliness of satellite remote sensing, great influence of cloud cover, low data interpretation accuracy, low efficiency of unmanned aerial vehicle cluster cooperation and task scheduling, insufficient fusion accuracy caused by time and space heterogeneity of multi-source data, low accuracy of dynamic change detection and tree species classification, single early warning mechanism and high false alarm rate, and lack of closed-loop linkage response lag.

[0004] To achieve the above-mentioned purpose, the present application provides the following technical scheme: a forestry resource dynamic monitoring system based on unmanned aerial vehicle patrol, comprising: An unmanned aerial vehicle patrol cluster module: comprising a master unmanned aerial vehicle and a slave unmanned aerial vehicle, the master unmanned aerial vehicle and the slave unmanned aerial vehicle are interconnected through a wireless communication network, the unmanned aerial vehicle is equipped with a multispectral imaging sensor, a laser radar (LiDAR), an infrared thermal imager and a positioning and navigation module, and is used for collecting real-time image data, terrain data and environmental parameters of forestry resources; A data fusion processing module: deployed on a ground server or in the cloud, used for receiving real-time data transmitted by the unmanned aerial vehicle patrol cluster module, fusing multi-source heterogeneous data (image data, point cloud data, environmental parameters) through a time and space registration algorithm, and generating standardized forestry resource basic data; A dynamic change analysis module: based on a deep learning model, comparing and analyzing the current forestry resource basic data with the reference data in the historical database, identifying the dynamic characteristics of forest coverage change, vegetation health status, topographic change and abnormal heat source, and generating a change characteristic map; An intelligent early warning decision module: pre-designing an abnormal threshold of forestry resources, automatically triggering an early warning mechanism when the data in the change characteristic map exceeds the threshold, generating an early warning position, type and level in combination with a geographic information system (GIS), and pushing to a management terminal; A task scheduling and cooperative control module: automatically planning the work path of the unmanned aerial vehicle patrol cluster according to the result of the dynamic change analysis module, optimizing the multi-unmanned aerial vehicle cooperative patrol strategy, and focusing on re-patrolling or data supplementing in the abnormal area.

[0005] Preferably, the cooperative patrol strategy of the unmanned aerial vehicle patrol cluster module comprises: Based on an improved ant colony algorithm or reinforcement learning algorithm, an optimal flight path of multiple unmanned aerial vehicles covering the forestry area is generated, ensuring that the patrol coverage rate is greater than or equal to 95% and the data collection overlap rate is less than or equal to 30%; The master unmanned aerial vehicle dynamically allocates patrol sub-tasks according to the power, data transmission quality and task area priority of each slave unmanned aerial vehicle, and adjusts the flight parameters in real time.

[0006] By means of the above technical solution, the track is dynamically optimized by an intelligent algorithm, compared with the traditional preset route operation, the inspection coverage rate is increased from less than 90% to more than 95%, the data overlap rate is reduced from more than 40% to less than 30%, the blind area of patrol is effectively eliminated and the power consumption of the unmanned aerial vehicle is reduced, and meanwhile, the task allocation mechanism based on real-time state can automatically allocate tasks when the power of the unmanned aerial vehicle is insufficient or the link is interrupted, thereby ensuring the continuity of monitoring.

[0007] Preferably, the data fusion processing module comprises: A multi-sensor time synchronization unit: the multi-spectral image, LiDAR point cloud and environmental parameters are time-stamped calibrated by a Beidou / GPS clock, and the error is less than or equal to 10 ms; A spatial coordinate conversion unit: different sensor data are uniformly mapped to a WGS84 coordinate system or a local projection coordinate system, and sub-pixel level registration is realized by a feature point matching algorithm; A data denoising and completion unit: wavelet transform or a generative adversarial network (GAN) is used to filter noise data and interpolate missing data.

[0008] By means of the above technical solution, sub-millisecond time synchronization and sub-pixel spatial registration of multi-source data are realized, the time error is less than or equal to 10 ms, and the spatial registration accuracy is less than or equal to 0.5 pixels, thereby solving the spatio-temporal misalignment problem of multi-spectral images and LiDAR point clouds, the reliability of the data after denoising and completion processing is increased to more than 85%, and a high-precision data basis is provided for single-tree level disease identification and terrain micro-change monitoring.

[0009] Preferably, the deep learning model of the dynamic change analysis module comprises: A convolutional neural network (CNN) for tree species identification, which classifies tree species based on a transfer learning technique, and the accuracy is greater than or equal to 92%; A time series feature extraction network for vegetation health assessment, which identifies disease and pest areas or drought stress areas by analyzing the time series changes of normalized difference vegetation index (NDVI) and leaf area index (LAI); A point cloud registration network for terrain change detection, which detects terrain anomalies by comparing LiDAR point cloud data at different times, and the minimum detection accuracy is less than or equal to 0.5 m 3 .

[0010] By means of the above technical solution, the tree species classification accuracy is increased from less than or equal to 85% of the traditional machine learning to 92%, more than 20 mixed forest tree species can be accurately distinguished, the early disease and pest identification period of the vegetation health assessment is more than 15 days, and the terrain anomaly detection accuracy is less than or equal to 0.5 m 3 , which meets the real-time monitoring requirements of landslides, illegal stone mining and other micro-terrain changes.

[0011] Preferably, the pre-warning types of the intelligent pre-warning decision module include: Forest resource destruction pre-warning: based on image semantic segmentation, illegal felling and forest land occupation behaviors are identified, with a positioning accuracy of ≤10 m; Forest fire pre-warning: through infrared thermal imaging and multispectral data fusion, abnormal heat sources with a temperature of ≥80°C are identified, and the fire spread trend is predicted in combination with wind direction data; Disease and pest pre-warning: based on vegetation spectral feature anomalies (such as red edge shift) and historical disease and pest databases, high-risk areas of disease and pests are predicted.

[0012] By adopting the above technical solution, the positioning accuracy of forest destruction is ≤10 m, which is 20 times more efficient than manual inspection; the fire pre-warning combines thermal imaging and wind direction data to predict the fire spread path in the next 30 minutes, with a false alarm rate reduced from 25% to below 8%; the disease and pest pre-warning combines spectral features and historical data to predict high-risk areas with an accuracy of ≥90%, triggering prevention and control responses 7-10 days in advance.

[0013] Preferably, the system further comprises a forestry resource digital twin module: Based on the fused multi-source data, a three-dimensional digital twin model of the forestry area is constructed to real-time map the dynamic process of forest growth and environmental changes; Through the digital twin model, the evolution trend of forestry resources under different climate conditions or human intervention is simulated to provide decision support for forestry management.

[0014] By adopting the above technical solution, a three-dimensional digital twin model with a construction accuracy of centimeter level is constructed to real-time map the forest growth (diameter at breast height, tree height annual change error ≤3%) and environmental evolution process, and through the simulation of forestry resource responses under different climate scenarios (such as extreme precipitation and high temperature), quantitative decision support is provided for carbon sink accounting and ecological restoration, with a carbon storage estimation error of ≤5%.

[0015] Preferably, the task scheduling and collaborative control module further comprises: Emergency response sub-module: when receiving the alarm of the intelligent pre-warning decision module, an unmanned aerial vehicle emergency inspection task is automatically generated, and the nearest unmanned aerial vehicle cluster is dispatched to arrive at the pre-warning area within 10 minutes; Energy consumption optimization sub-module: in combination with the remaining battery capacity of the unmanned aerial vehicle and the priority of the inspection task, a return path with optimal energy consumption is generated through a mixed integer programming algorithm.

[0016] By adopting the above technical solution, the emergency response time is compressed from more than 30 minutes in traditional manual scheduling to within 10 minutes, the nearest unmanned aerial vehicle cluster can quickly arrive at the pre-warning area, the energy consumption optimization algorithm reduces the energy consumption of the unmanned aerial vehicle return path by 20%, prolongs the endurance time by 30 minutes, and improves the continuous monitoring capability in complex terrain.

[0017] Preferably, the monitoring process of this monitoring system includes the following steps: Data acquisition steps: Collect multi-source data of the forestry area through the drone patrol cluster module. The data includes multispectral images, LiDAR point clouds, temperature and humidity parameters, and GPS trajectory. Fusion processing steps: Spatiotemporal registration and noise reduction are performed on multi-source data to generate a standardized basic dataset of forestry resources; Dynamic analysis steps: Based on a deep learning model, compare current data with historical data to identify the dynamic changes in forestry resources; Early warning decision-making steps: Trigger an early warning based on the characteristics of the change, and generate handling suggestions for the abnormal area; Task scheduling steps: Based on the early warning results and data analysis requirements, optimize the patrol tasks and collaborative strategies of multiple drones.

[0018] By adopting the above technical solution, a fully automated closed-loop process of "collection-fusion-analysis-early warning-schedule" is formed, which improves efficiency by more than 10 times compared with the traditional manual intervention mode, and the data processing delay is ≤5 minutes, realizing the transformation from passive monitoring to proactive early warning and meeting the needs of dynamic supervision across the entire domain.

[0019] Preferably, the method for detecting changes in forest cover in the dynamic analysis step includes: By using the difference operation and threshold segmentation of multi-temporal images, the newly added or reduced forest cover area can be extracted; By combining forest resource archive data, the accuracy of the detection results was verified to be ≥90% through a confusion matrix.

[0020] By adopting the above technical solution and combining multi-temporal imagery with forest archive data, the accuracy rate of forest cover change detection is ≥90%, which can identify more than 0.5 hectares of felled or newly added forest land, which is 4 times higher than satellite remote sensing (identification accuracy ≥2 hectares), effectively curbing covert illegal logging.

[0021] Preferably, in the task scheduling step, the path planning method for multi-UAV collaborative patrol includes: Establish a gridded task map for forestry areas, with the inspection priority of each grid determined by the frequency of historical anomalies and its ecological importance; An improved particle swarm optimization algorithm is adopted to maximize the inspection value density per unit time while satisfying the drone's endurance constraint.

[0022] By adopting the above technical solution, through grid-based priority division and improved algorithm optimization, the inspection value density per unit time is increased by more than 50%, the repetitive inspection frequency of high-priority areas (such as ecologically sensitive areas and areas with a high incidence of historical anomalies) is increased by 3 times, and the drone endurance constraints are met, thereby maximizing monitoring efficiency and resource utilization.

[0023] Compared with the prior art, the beneficial effects of the present invention are: the forestry resource dynamic monitoring system based on drone patrol: 1. Through dynamic task allocation strategies and real-time anomaly response mechanisms, when the drone's battery is low or data is interrupted, the system automatically triggers task reassignment, eliminating patrol blind spots. Compared with traditional single-flight or preset route operations, the inspection coverage rate has increased from less than 90% to more than 98%, and the data overlap rate has decreased from more than 40% to less than 15%, significantly reducing resource waste and monitoring omission risks. For abnormal events such as sudden fires, the system dynamically adjusts the inspection priority based on real-time data, reducing the emergency response time from more than 30 minutes to within 5 minutes, meeting the real-time capture needs of the "golden period" in the early stages of disasters, and effectively improving the timeliness of disaster prevention and control. 2. By using a unified clock synchronization protocol and a sub-meter-level spatial registration algorithm, the time synchronization error has been reduced from more than 100ms to less than 10ms, and the spatial registration error in complex terrain areas has been optimized from several meters to less than 0.5 meters. This achieves accurate spatiotemporal matching of spectral data and point cloud data. For low-density point clouds and noisy images, deep learning interpolation and denoising algorithms are used to improve the data restoration reliability from less than 60% to more than 85%. This provides high-precision data support for single-tree-level disease identification and small-scale logging monitoring, solving the problem of insufficient reliability of traditional interpolation methods. 3. By using deep learning algorithms (such as convolutional neural networks), the accuracy of forest cover change identification has been improved from 70%-85% to over 95%, accurately distinguishing between natural vegetation decay and human logging. The accuracy of tree species classification in mixed forest areas has been improved from less than 85% to 92%, meeting the needs of refined management. By integrating multi-source data such as spectral features, terrain parameters, and temperature to build an early warning model, the false alarm rate has been reduced from over 25% to below 8%. After the early warning is triggered, the patrol equipment is automatically linked to generate a re-patrol task, realizing closed-loop management of "early warning-handling-verification". The response lag time has been shortened from several hours of manual planning to real-time scheduling. 4. By acquiring and fusing high-precision data, the system enables 3D modeling and dynamic updating of real-world forest areas. This significantly reduces the technical difficulty of incorporating the coverage rate of digital twin forest farms into carbon sequestration assessment indicators, providing a data foundation for accurate accounting of forestry carbon sequestration capacity. At the same time, the system uses real-time perception, intelligent analysis, and prediction models to identify potential disaster risks in advance (such as trends in pest and disease outbreaks and fire risk areas), promoting the transformation of forestry resource management from post-event handling to pre-event prevention, and meeting the strategic needs of national ecological security and sustainable development. 5. It integrates "collection-fusion-analysis-early warning-dispatch" into one system, is compatible with existing satellite remote sensing and ground station data, and forms a three-dimensional monitoring network covering air, land and space. It solves the limitations of traditional "point-to-line" level monitoring, provides efficient and accurate technical support for tasks such as forest disaster prevention and control and ecological restoration, and significantly improves the response efficiency and scientific decision-making of smart forestry management. Attached Figure Description

[0024] Fig. 1 This is a schematic diagram of the system composition of the present invention; Fig. 2 This is a schematic diagram of the monitoring process of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] Please see Figs. 1-2 This invention provides a technical solution: a dynamic monitoring system for forestry resources based on drone patrols.

[0027] Example 1: Data Acquisition and Fusion Processing Based on UAV Swarm Implementation scenario: Routine patrols of a mountainous forest area (approximately 10,000 hectares). Specific steps: Data Acquisition: A cluster of 5 drones is deployed, equipped with multispectral cameras (0.5-meter resolution), LiDAR (50 points / square meter point cloud density), and environmental sensors, to conduct inspections according to a dynamic task allocation strategy. Each sortie covers approximately 2,000 hectares, completing a full-area scan in a single day, and simultaneously recording GPS tracks (positioning accuracy down to sub-meter level).

[0028] Fusion Processing: Multi-source data is spatiotemporally aligned using a unified clock synchronization protocol (error ≤10ms) and a deep learning-based spatial registration algorithm (error ≤0.3 meters in complex terrain areas). A denoising algorithm combining wavelet transform and GAN network is employed to repair low-density point cloud areas, improving data reliability to 88%. Technical Results: Compared to traditional manual patrols (daily coverage <500 hectares, data error rate >20%), inspection efficiency is increased by 4 times, and data fusion accuracy meets the requirements for single-tree-level lesion identification (e.g., early symptom identification rate of pine wilt disease ≥90%).

[0029] Example 2: Detection and Dynamic Analysis of Forest Cover Changes Implementation scenario: Monitoring of illegal logging in nature reserves. Specific steps: Change detection: Multi-temporal images of the area in May and August 2025 were acquired. New logging areas (area ≥ 0.5 hectares) were extracted by difference calculation (band combination R865 / R550) and Otsu threshold segmentation. Combined with forest distribution data in the forest resource archive, the detection results were verified by confusion matrix, with an accuracy of 92% (compared to only 80% by the traditional threshold segmentation method).

[0030] Feature Recognition: Using a convolutional neural network (U-Net model), the system distinguishes between natural decay and human logging. Testing on 100 samples showed a classification accuracy of 95%, with a 15% improvement in the recognition accuracy of human features such as serrated tree stump edges and regular logging areas. Technical Results: Three concealed logging sites (0.8-1.2 hectares) were successfully identified, detecting anomalies 15 days earlier than satellite remote sensing (10-meter resolution, difficult to identify areas <2 hectares), preventing forest loss of over 200 cubic meters.

[0031] Example 3: Multi-UAV Cooperative Task Scheduling and Path Optimization Implementation scenario: Emergency response to sudden forest fires. Specific steps: Gridded task map: The forest area is divided into 100m×100m grids. Combining historical fire frequency (distribution of fire points in the past 3 years) and ecological importance (such as the distribution area of ​​primary forest), 200 high-priority grids (weight ≥ 0.8) are set.

[0032] Path planning: An improved particle swarm optimization algorithm (introducing adaptive inertial weights and an elite back-learning strategy) was used to assign tasks to 8 UAVs. Constraints included: single-UAV battery power supporting 2 hours of flight (covering approximately 150 grids) and real-time data transmission link connectivity. After optimization, the inspection value density per unit time (unit: effective anomaly identifications / hour) increased by 60%, and the data overlap rate decreased to 12%. Technical results: Fire detection time was reduced from 28 minutes on traditional preset routes to 4 minutes and 30 seconds. Subsequent re-inspection tasks were automatically generated, forming a "warning-response-verification" closed loop, improving efficiency by 8 times compared to manual scheduling.

[0033] Example 4: Integrated Application of Digital Twin and Carbon Sequestration Implementation scenario: Quarterly carbon storage assessment of carbon sink forests. Specific steps: 3D modeling: By fusing UAV LiDAR point cloud (point cloud density 80 points / square meter) with multispectral data, a digital twin model of the forest area is constructed to achieve accurate measurement of individual tree diameter at breast height and tree height (error ≤3%).

[0034] Carbon sink analysis: Combining forest growth models (such as allometric growth equations), the system calculates changes in carbon storage in the tree layer. Compared with traditional plot survey methods (error >15%), the system achieves a carbon storage accuracy of 94% and improves coverage efficiency by 20 times. Technical results: The system successfully incorporates the coverage rate (100%) of the digital twin forest farm into the carbon sink assessment indicators, providing accurate accounting data for a forestry carbon sink project, verifying carbon emission reduction errors of <5%, and meeting international voluntary carbon market (VCS) standards.

[0035] Example 5: Pest and Disease Early Warning Based on Multi-Source Data Fusion Implementation scenario: Early warning of pine caterpillar outbreak in pine forests. Specific steps: Multi-dimensional data fusion: Integrating UAV multispectral data (vegetation index NDVI, PRI), ground station temperature and humidity data (accuracy ±0.5℃, ±2%RH) and historical pest and disease records to construct a random forest early warning model.

[0036] Intelligent Early Warning: An early warning is triggered when the model outputs a risk probability ≥ 70%. In its application in June 2025, it predicted three high-incidence areas of pine caterpillars 10 days in advance (traditional single-indicator early warnings only predict 3 days in advance), reducing the false alarm rate from 25% to 7%. Technical Results: By spraying biological pesticides in advance, pest control costs were reduced by 40%, and the affected area of ​​pine forests was reduced by 85%, validating the effectiveness of multi-source data fusion early warning.

[0037] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A forestry resource dynamic monitoring system based on unmanned aerial vehicle (UAV) patrols, characterized in that: The system includes: The drone patrol cluster module includes a master drone and slave drones, which are interconnected via a wireless communication network. Each drone is equipped with a multispectral imaging sensor, LiDAR, infrared thermal imager, and positioning and navigation module to collect real-time image data, terrain data, and environmental parameters of forestry resources. Data fusion processing module: Deployed on a ground server or in the cloud, it receives real-time data transmitted by the UAV patrol cluster module and uses a spatiotemporal registration algorithm to fuse multi-source heterogeneous data (image data, point cloud data, environmental parameters) to generate standardized basic forestry resource data. Dynamic Change Analysis Module: Based on a deep learning model, it compares and analyzes the current basic data of forestry resources with the benchmark data in the historical database to identify the dynamic characteristics of changes in forest cover, vegetation health status, topographic and geomorphological changes and abnormal heat sources, and generates a change feature map. Intelligent early warning decision module: It pre-designs forestry resource anomaly thresholds. When the data in the change feature map exceeds the threshold, it automatically triggers the early warning mechanism, combines the geographic information system (GIS) to generate the early warning location, type and level, and pushes it to the management terminal; Task scheduling and collaborative control module: Based on the results of the dynamic change analysis module, automatically plan the operation path of the UAV patrol cluster, optimize the multi-UAV collaborative patrol strategy, and conduct key re-patrols or data supplementation in abnormal areas.

2. The forestry resource dynamic monitoring system based on unmanned aerial vehicle (UAV) patrol as described in claim 1, characterized in that: The collaborative patrol strategy of the drone patrol cluster module includes: Based on improved ant colony algorithms or reinforcement learning algorithms, optimal flight paths for multiple drones to cover forestry areas are generated, ensuring that the inspection coverage rate is ≥95% and the data collection overlap rate is ≤30%. The master control UAV dynamically allocates inspection sub-tasks based on the battery level, data transmission quality, and task area priority of each slave control UAV, and adjusts flight parameters in real time.

3. The forestry resource dynamic monitoring system based on unmanned aerial vehicle (UAV) patrol as described in claim 1, characterized in that: The data fusion processing module includes: Multi-sensor time synchronization unit: timestamps multispectral images, LiDAR point clouds and environmental parameters using BeiDou / GPS clocks, with an error ≤10ms; Spatial coordinate transformation unit: maps data from different sensors to the WGS84 coordinate system or the local projected coordinate system, and achieves sub-pixel level registration through a feature point matching algorithm; Data denoising and completion unit: Wavelet transform or generative adversarial network (GAN) is used to filter noisy data and interpolate to complete missing data.

4. The forestry resource dynamic monitoring system based on unmanned aerial vehicle (UAV) patrol as described in claim 1, characterized in that: The deep learning model of the dynamic change analysis module includes: A convolutional neural network (CNN) for tree species identification, based on transfer learning techniques, achieves an accuracy of ≥92% in classifying tree species. A temporal feature extraction network for vegetation health assessment identifies areas under pest, disease, or drought stress by analyzing temporal changes in the Normalized Difference Vegetation Index (NDVI) and Leaf Area Index (LAI). A point cloud registration network for terrain change detection detects terrain anomalies by comparing LiDAR point cloud data from different periods, with a minimum detection accuracy of ≤0.5m. 3 .

5. The forestry resource dynamic monitoring system based on unmanned aerial vehicle (UAV) patrol as described in claim 1, characterized in that: The warning types of the intelligent early warning decision module include: Early warning of forest resource damage: Identification of illegal logging and forest land encroachment based on image semantic segmentation, with a positioning accuracy of ≤10m; Forest fire early warning: By fusing infrared thermal imaging with multispectral data, abnormal heat sources with a temperature ≥80℃ are identified and fire spread trends are predicted by combining wind direction data; Pest and disease early warning: Based on abnormal vegetation spectral characteristics (such as red edge shift) and historical pest and disease databases, predict areas with high incidence of pests and diseases.

6. The forestry resource dynamic monitoring system based on unmanned aerial vehicle (UAV) patrol as described in claim 1, characterized in that: The system also includes a digital twin module for forestry resources: A three-dimensional digital twin model of the forestry area is constructed based on the fused multi-source data to map the dynamic process of tree growth and environmental change in real time. By simulating the evolution trends of forestry resources under different climatic conditions or human intervention through digital twin models, decision support can be provided for forestry management.

7. The forestry resource dynamic monitoring system based on unmanned aerial vehicle (UAV) patrol as described in claim 1, characterized in that: The task scheduling and collaborative control module also includes: Emergency Response Submodule: When receiving an alarm from the intelligent early warning decision module, it automatically generates an emergency drone inspection task and dispatches the nearest drone cluster to arrive at the warning area within 10 minutes; Energy consumption optimization submodule: Combining the remaining power of the drone battery with the priority of the inspection task, it generates the energy-optimal return path through a mixed integer programming algorithm.

8. The forestry resource dynamic monitoring system based on unmanned aerial vehicle (UAV) patrol as described in claim 1, characterized in that: The monitoring process of this monitoring system includes the following steps: Data acquisition steps: Collect multi-source data of the forestry area through the drone patrol cluster module. The data includes multispectral images, LiDAR point clouds, temperature and humidity parameters, and GPS trajectory. Fusion processing steps: Spatiotemporal registration and noise reduction are performed on multi-source data to generate a standardized basic dataset of forestry resources; Dynamic analysis steps: Based on a deep learning model, compare current data with historical data to identify the dynamic changes in forestry resources; Early warning decision-making steps: Trigger an early warning based on the characteristics of the change, and generate handling suggestions for the abnormal area; Task scheduling steps: Based on the early warning results and data analysis requirements, optimize the patrol tasks and collaborative strategies of multiple drones.

9. The forestry resource dynamic monitoring system based on unmanned aerial vehicle (UAV) patrol as described in claim 8, characterized in that: The dynamic analysis step includes methods for detecting changes in forest cover, including: By using the difference operation and threshold segmentation of multi-temporal images, the newly added or reduced forest cover area can be extracted; By combining forest resource archive data, the accuracy of the detection results was verified to be ≥90% through a confusion matrix.

10. The forestry resource dynamic monitoring system based on unmanned aerial vehicle (UAV) patrol as described in claim 8, characterized in that: The path planning method for multi-UAV collaborative patrol in the task scheduling step includes: Establish a gridded task map for forestry areas, with the inspection priority of each grid determined by the frequency of historical anomalies and its ecological importance; An improved particle swarm optimization algorithm is adopted to maximize the inspection value density per unit time while satisfying the drone's endurance constraint.

Images