Forest degradation monitoring and evaluating method and system based on machine vision

Abstract

The invention discloses a forest degradation monitoring and evaluation method and system based on machine vision, and relates to the technical field of information. Comprising a multi-source data acquisition module, a multi-dimensional dynamic degradation index calculation module, an environment situation perception module, a monitoring strategy self-evolution engine module, a sensor mode switching module, an edge-cloud collaborative calculation module and a visual decision terminal module. The system has the advantages that the problem of poor assessment capability in the prior art is solved by integrating a plurality of functional modules such as the multi-source data acquisition module and the multi-dimensional dynamic degradation index calculation module, the multi-source data acquisition module acquires data in multiple aspects such as vegetation, soil, climate and human activities in real time, rich information is provided for comprehensive assessment, and the assessment efficiency is improved. And the multi-dimensional dynamic degradation index calculation module integrates the data, so that accurate evaluation and differentiated treatment are realized, and the comprehensiveness and accuracy of forest degradation evaluation are greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a forest degradation monitoring and assessment method and system based on machine vision. Background Art

[0002] As an important part of the earth's ecosystem, forests play an irreplaceable role in maintaining ecological balance, providing ecological services and promoting economic development. With the intensification of global climate change and human activities, the problem of forest degradation is becoming increasingly serious, bringing many challenges to the ecological environment and human society. Accurate monitoring and assessment of forest degradation is crucial to the formulation of scientific and reasonable forest protection and restoration strategies. Traditional forest monitoring methods, such as field surveys, not only consume a lot of manpower, material resources and time, but also have a limited monitoring range, making it difficult to achieve real-time and dynamic monitoring of large areas of forests.

[0003] There are certain defects in the existing technology. The first is that the existing technology has poor assessment capabilities. The existing forest degradation assessment system relies on a single indicator (such as the NDVI vegetation index) or a static weight model, and cannot dynamically integrate multi-dimensional data of vegetation, soil, climate and human activities, resulting in one-sided assessment results and difficulty in providing a scientific basis for differentiated governance. Secondly, the existing technology lacks dynamic adaptability. The existing system cannot quickly adjust the monitoring strategy when the environment changes suddenly (extreme climate, sudden insect pests), resulting in failure of data collection and model analysis. When heavy rains suddenly occur, drone inspections are still carried out as planned, and the image quality is low. The frequency of thermal infrared monitoring is not increased in a targeted manner during the insect pest outbreak period, and the missed detection rate is high. To this end, we propose a forest degradation monitoring and assessment method and system based on machine vision. Summary of the invention

[0004] The purpose of the present invention is to provide a forest degradation monitoring and assessment method and system based on machine vision.

[0005] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a forest degradation monitoring and assessment system based on machine vision, the system comprising a multi-source data acquisition module, a multi-dimensional dynamic degradation index calculation module, an environmental situation awareness module, a monitoring strategy self-evolution engine module, a sensor mode switching module, an edge-cloud collaborative computing module and a visual decision terminal module;

[0006] Multi-source data acquisition module: Integrates a drone multi-spectral camera, a ground capacitive soil sensor, a meteorological API interface, and a sound recognition unit. The vegetation image data collected by the drone multi-spectral camera is transmitted in real time to the drone terminal in the multi-dimensional dynamic degradation index calculation module and the edge-cloud collaborative calculation module through a wireless transmission module. The soil moisture data measured by the ground capacitive soil sensor is transmitted to the multi-dimensional dynamic degradation index calculation module wirelessly. The meteorological API interface obtains meteorological data, including temperature, air pressure, rainfall, etc. On the one hand, it is transmitted to the environmental situation perception module for calculating the meteorological mutation index, and on the other hand, it is also transmitted to the multi-dimensional dynamic degradation index calculation module. The human activity sound data detected by the sound recognition unit is transmitted to the multi-dimensional dynamic degradation index calculation module for calculating the human activity density parameter, and is also transmitted to the monitoring strategy self-evolution engine module to provide a reference for formulating monitoring strategies. The multi-source data acquisition module obtains vegetation index, soil moisture, rainfall, and human activity data in real time;

[0007] Multi-dimensional dynamic degradation index calculation module: Receives the data transmitted from the multi-source data acquisition module and calculates the dynamic degradation index based on the formula. The specific calculation formula is as follows:

[0008]

[0009] Where: MDDI represents the degradation index, and λ, μ, v, and ξ are adaptively allocated according to the forest area type through the Analytic Hierarchy Process (AHP), and λ + μ + ν + ξ = 1. The multi-dimensional dynamic degradation index calculation module transmits the calculated data to the visualization decision terminal module for generating a degradation level heat map and governance suggestions. On the other hand, it is transmitted to the LSTM network in the cloud as one of the input data for predicting the degradation trend;

[0010] The specific steps of the Analytic Hierarchy Process (AHP) are as follows:

[0011] Construct a hierarchical model: Take the forest degradation influencing factors (vegetation, soil, climate, human activities) as the criterion layer;

[0012] Construct a judgment matrix: Compare the pairwise importance of the criterion layer factors according to expert experience or historical data;

[0013] Calculate the weights: Determine the weight coefficients λ, μ, v, and ξ of each criterion through the eigenvector method and verify the consistency (CR < 0.1);

[0014] Dynamic adjustment: Preset a weight template according to the forest area type (such as tropical rainforest, arid forest area), and fine-tune it in combination with real-time data;

[0015] Environmental Situation Awareness Module: It calculates the meteorological mutation index CMI and pest density in real time through a weather station and a bionic vibration sensor. The changing data of meteorological parameters such as temperature and air pressure collected by the weather station are used to calculate the meteorological mutation index, and the calculation result is transmitted to the monitoring strategy self-evolution engine module. The characteristic vibration signals in the range of 200Hz - 500Hz detected by the bionic vibration sensor are processed and calculated to obtain pest density data, which is also transmitted to the monitoring strategy self-evolution engine module to provide environmental situation information for the dynamic generation of monitoring strategies;

[0016] Monitoring Strategy Self-evolution Engine Module: It receives the meteorological mutation index and pest density data transmitted by the environmental situation awareness module, generates dynamic task scheduling instructions based on these data. When the meteorological mutation index exceeds the threshold or the pest density reaches a certain level, it calculates and generates scheduling instructions for the UAV cluster according to a specific formula, and transmits them to the UAV terminal to control the execution of the UAV monitoring tasks. At the same time, it will also trigger the operation of the sensor modality switching module according to the data situation;

[0017] Sensor Modality Switching Module: After receiving the trigger instruction from the monitoring strategy self-evolution engine module, it enables the synthetic aperture radar in extreme weather, and the data collected by it is transmitted to the edge-cloud collaborative computing module for processing. When pests break out, it switches to a thermal infrared imager, and the collected data is also transmitted to the edge-cloud collaborative computing module for subsequent analysis and processing;

[0018] Edge-Cloud Collaborative Computing Module: It deploys a lightweight model at the UAV terminal, receives the data transmitted by the UAV multi-spectral camera, synthetic aperture radar and thermal infrared imager, and realizes real-time segmentation of vegetation coverage. On the one hand, the segmentation result is fed back to the monitoring strategy self-evolution engine module to provide a basis for further adjustment of monitoring strategies. On the other hand, it is transmitted to the LSTM network in the cloud, receives the MDDI historical sequence from the multi-dimensional dynamic degradation index calculation module, meteorological forecast data and the segmentation result data transmitted by the UAV terminal, predicts the degradation trend, and transmits the prediction result to the visualization decision terminal module;

[0019] Visualization Decision Terminal Module: It receives the degradation index transmitted by the multi-dimensional dynamic degradation index calculation module and the degradation trend prediction result transmitted by the edge-cloud collaborative computing module, maps the degradation index value into a heat map, and matches the governance strategy according to the degradation level and cause type (pest / deforestation / fire), and generates a multi-dimensional report including the replanting area and law enforcement priority.

[0020] As a further solution of the present invention: In the multi-dimensional dynamic degradation index calculation module, the EVI baselineDenotes the baseline enhanced vegetation index, which is the enhanced vegetation index value of a certain forest area in a relatively stable state in the past obtained through historical database calls. After matching with the current GPS coordinates, it serves as a comparison benchmark to measure the change of the vegetation index at the current moment. The vegetation index is an important indicator to measure the growth status and coverage of vegetation. Compared with other vegetation indices, the enhanced vegetation index can better eliminate the influence of noise such as atmosphere and soil background and more accurately reflect the true state of vegetation. The EVI current Denotes the enhanced vegetation index at the current moment, which is obtained by collecting the current vegetation image data through the drone multispectral camera in the multi-source data acquisition module and after processing and analysis, and is used to compare with the EVI baseline , to reflect the real-time change of the vegetation growth status. The SM dry Denotes the soil moisture in the dry state, which is the soil moisture data measured by the ground capacitive soil sensor during the time period with a monthly time cycle and the soil moisture continuously lower than 60% of the historical average soil moisture of this forest area within this cycle. The SM wet , denotes the soil moisture in the wet state, which is the data obtained by measuring through the ground capacitive soil sensor during the time period with a monthly time cycle and the soil moisture continuously higher than 80% of the historical average soil moisture of this forest area within this cycle. The RF anomaly Denotes the rainfall anomaly value, which is the difference obtained by comparing the current rainfall data obtained through the meteorological API interface with the average rainfall in the same period of history in this area. The RF avg Denotes the average rainfall in the same period of history in this area, which is obtained based on long-term meteorological data statistics and analysis and serves as a reference standard to measure whether the current rainfall is abnormal. The Denotes the density of the human activity environment situation perception module, which is obtained by detecting the sound pattern of logging operations through the sound recognition unit and combining with the YOLOv7 model to identify fire points and vehicle targets and then making statistics.

[0021] As a further solution of the present invention: The environment situation perception module includes:

[0022] The calculation formula for the meteorological mutation index CMI is as follows:

[0023]

[0024] Among them, ΔT represents the average daily cycle temperature change rate, and ΔP represents the average daily cycle air pressure change rate. The data source is the distributed meteorological station network;

[0025] The pest density Pest density Detects the 200Hz - 500Hz characteristic vibration signal through the bionic vibration sensor and calculates the peak number per unit time.

[0026] As a further solution of the present invention: The monitoring strategy self-evolution engine module performs the following operations:

[0027] When the CMI exceeds the set threshold, calculate the priority weight of the UAV to perform the monitoring task according to the formula. The specific calculation formula is as follows:

[0028] W drone = α·CMI + β·Pest density

[0029] Where α and β are weight coefficients optimized online through reinforcement learning. According to the calculated W drone Generate a scheduling instruction for the UAV cluster, transmit it to the UAV terminal, and control the execution of the UAV's monitoring task;

[0030] When Pest density > 5 times / minute, trigger the collaborative positioning of the thermal infrared camera and the vibration sensor for the core area of the pest damage.

[0031] As a further solution of the present invention: In the edge-cloud collaborative computing module, a lightweight U-Net model is deployed on the UAV terminal. The lightweight U-Net model is processed by model compression and quantization optimization techniques to reduce the consumption of computing resources while ensuring the segmentation accuracy. The cloud uses an LSTM network to predict the degradation trend, and the input data includes the MDDI historical sequence and the weather forecast.

[0032] As a further solution of the present invention: In the visual decision terminal module, the visual decision terminal module maps the degradation index value into a three-color heat map of red (severe), orange (moderate), and green (mild). The mapping process uses a threshold division method, and according to the degradation level and the cause type (pest damage / deforestation / fire), matches the governance strategy, generates a multi-dimensional report including the replanting area and the law enforcement priority, and the report generation process combines historical governance data and expert experience.

[0033] As a further solution of the present invention: After the monitoring strategy self-evolution engine module calculates W drone it will generate corresponding scheduling instructions according to different threshold ranges:

[0034] W drone ≥ 8 (high priority): Immediately deploy all available UAVs, require them to fly to the target area at the fastest speed and the shortest path, collect images at the highest resolution, and transmit the monitoring data in real time every 5 minutes to conduct a comprehensive and high-frequency monitoring of the forest environment without dead angles;

[0035] 5 ≤ W drone<8 (Medium priority): Arrange two-thirds of the drone swarm to conduct detailed monitoring of key areas according to the pre-planned partitions. These drones need to maintain medium-resolution image acquisition and transmit data every 15 minutes to timely grasp the dynamic changes of the forest;

[0036] 2 ≤ W drone <5 (Low priority): Dispatch one-quarter of the drones to perform routine patrol tasks, fly at normal speed and along the default path, use low-resolution image acquisition, and transmit data every 30 minutes to maintain basic monitoring of the forest conditions;

[0037] W drone <2 (Very low priority): Only send out individual drones for overview patrols, fly at a slightly slower speed, with lower-resolution image acquisition, and transmit data every hour, mainly to confirm whether there are any abnormal situations in the overall forest.

[0038] As a further aspect of the present invention: The EVI current is obtained through the following processing and analysis process: The current vegetation image data collected by the drone multispectral camera is first radiometrically calibrated to convert the digital quantization value obtained by the camera into an absolute radiance value, eliminating the radiation error caused by the camera response difference, ensuring the comparability of the data collected at different times and under different conditions. Subsequently, using the atmospheric correction algorithm, the influence of atmospheric molecular scattering, absorption, and aerosol scattering on light is removed to restore the true reflectance information of the ground objects. Then, image mosaicking and cropping techniques are used to stitch multiple collected images into a complete large-area image and crop it according to the boundary of the study area to obtain the target area image. After that, a supervised classification algorithm is used to classify the pixels in the image into vegetation and non-vegetation categories based on the known spectral characteristics of the vegetation samples, and the vegetation area is extracted. Finally, within the vegetation area, EVI is calculated according to the formula current EVI current The specific calculation formula is as follows:

[0039]

[0040] where ρ NIR ρ RED and ρ Blue represent the reflectance of the near-infrared band, red band, and blue band, respectively.

[0041] As a further aspect of the present invention: When the LSTM network in the cloud processes the input data for degradation trend prediction, it performs the following operations:

[0042] The MDDI historical sequence data received from the multi-dimensional dynamic degradation index calculation module is first normalized to map the data to the interval [0, 1] to eliminate the influence of data dimension differences on model training. Then, using the sliding window technique, the MDDI historical sequence is divided into time-step data blocks of a fixed length, and each data block is used as an input sample for the LSTM network;

[0043] For the data from weather forecasts, it is aligned and matched with the MDDI historical sequence data in the time dimension to ensure the temporal consistency of the data. Different variables in the weather forecast data, such as temperature, humidity, and rainfall, are respectively subjected to feature engineering to extract the features related to the forest degradation trend.

[0044] The vegetation cover segmentation results transmitted from the drone terminal in the edge-cloud collaborative computing module are converted into feature vectors that can characterize the changes in forest vegetation cover, such as the change ratio of vegetation cover area, the fragmentation degree of vegetation patches, etc.;

[0045] The processed MDDI historical sequence data, weather forecast feature data, and vegetation cover change feature vectors are concatenated to form a complete input data set. This data set is divided into a training set, a validation set, and a test set according to a certain ratio. The training set is used to train the LSTM network model, and the weights and parameters of the model are continuously adjusted through the backpropagation algorithm to minimize the error between the predicted value and the true value. The validation set is used to evaluate the performance of the model during training to prevent overfitting of the model. The test set is used to perform the final performance test on the trained model, and evaluation metrics such as the prediction accuracy rate and mean square error of the model are calculated;

[0046] The trained LSTM network model predicts the forest degradation trend based on the input weather forecast data at the current moment and for a period of time in the future, combined with the MDDI historical sequence features, and outputs the predicted values of the forest degradation degree for multiple future time periods, providing a decision-making basis for forest resource management.

[0047] In addition, the present invention also provides a method for monitoring and evaluating forest degradation based on machine vision, and the method includes the following steps:

[0048] Data collection steps: Use a multi-source data collection module integrated with a drone multi-spectral camera, a ground capacitive soil sensor, a meteorological API interface, and a sound recognition unit to collect data. Among them, the drone multi-spectral camera collects vegetation image data and sends it in real time to the drone terminal in the multi-dimensional dynamic degradation index calculation module and the edge-cloud collaborative calculation module through a wireless transmission module. The ground capacitive soil sensor measures soil humidity data and transmits it wirelessly to the multi-dimensional dynamic degradation index calculation module. The meteorological API interface obtains meteorological data and transmits it to the environmental situation perception module for calculating the meteorological mutation index and the multi-dimensional dynamic degradation index calculation module respectively. The sound recognition unit detects human activity sound data and transmits it to the multi-dimensional dynamic degradation index calculation module for calculating the human activity density parameter, and at the same time transmits it to the monitoring strategy self-evolution engine module to achieve real-time acquisition of vegetation index, soil humidity, rainfall, and human activity data;

[0049] Data calculation and analysis steps: The multi-dimensional dynamic degradation index calculation module receives the collected data and calculates the degradation index based on the formula The environmental situation perception module collects meteorological parameter change data through a meteorological station and calculates the meteorological mutation index using the formula At the same time, it detects the 200Hz - 500Hz characteristic vibration signal through a bionic vibration sensor to calculate the pest density, and transmits the meteorological mutation index and pest density data to the monitoring strategy self-evolution engine module;

[0050] Monitoring strategy formulation and execution steps: The monitoring strategy self-evolution engine module receives the data transmitted by the environmental situation perception module. When the meteorological mutation index exceeds the threshold or the pest density reaches a certain level, it calculates the priority weight of the drone to execute the monitoring task according to the formula W drone = α·CMI + β·Pest density Generates a scheduling instruction for the drone cluster according to the calculation result, transmits it to the drone terminal to control the execution of the monitoring task, and at the same time triggers the sensor mode switching module to work according to the data situation. In extreme weather, the sensor mode switching module enables a synthetic aperture radar, and switches to a thermal infrared imager during a pest outbreak. The collected data is transmitted to the edge-cloud collaborative calculation module. The drone terminal of the edge-cloud collaborative calculation module uses a lightweight U-Net model to perform real-time segmentation of the received data for vegetation coverage. On the one hand, the segmentation result is fed back to the monitoring strategy self-evolution engine module, and on the other hand, it is transmitted to the cloud LSTM network. The cloud LSTM network combines the MDDI historical sequence of the multi-dimensional dynamic degradation index calculation module and meteorological forecast data to predict the degradation trend;

[0051] Extreme weather is defined as:

[0052] Wind speed ≥ 15 m / s, rainfall ≥ 50 mm / h, or temperature mutation rate ΔT / Δt ≥ 5 °C / h;

[0053] The triggering conditions for pest outbreaks are:

[0054] Pest density ≥ 5 times / min, and lasting for more than 10 min;

[0055] Result presentation and decision-making steps: The visual decision terminal module receives the degradation index from the multi-dimensional dynamic degradation index calculation module and the degradation trend prediction result from the edge-cloud collaborative calculation module, maps the degradation index value into a three-color heat map of red (severe), orange (moderate), and green (mild) through the threshold division method, and matches the governance strategy according to the degradation level and cause type (pest / deforestation / fire), combining historical governance data and expert experience, generating a multi-dimensional report including replanting area and law enforcement priority, providing a decision-making basis for forest resource management.

[0056] Adopting the above technical solution, compared with the prior art, the beneficial effects of the present invention are as follows:

[0057] 1. By integrating multiple functional modules such as the multi-source data acquisition module and the multi-dimensional dynamic degradation index calculation module, the present invention solves the problem of poor evaluation ability in the prior art. The multi-source data acquisition module obtains real-time data on vegetation, soil, climate, and human activities in multiple aspects, providing rich information for comprehensive evaluation. The multi-dimensional dynamic degradation index calculation module integrates these data, abandoning the method in the prior art that relies on a single index or a static weight model. This module calculates the degradation index through a unique formula, comprehensively considering the influence of various factors on forest degradation, no longer limited to a single vegetation index. At the same time, the weight is automatically assigned according to the forest area type, making the evaluation result more in line with the actual situation of different forest areas. Based on these data and calculation results, the visual decision terminal module can map the degradation index value into a heat map and generate a multi-dimensional report including replanting area and law enforcement priority, providing a scientific basis for forest resource management, realizing precise evaluation and differential governance, and greatly improving the comprehensiveness and accuracy of forest degradation evaluation;

[0058] 2. Through the environmental situation awareness module and the monitoring strategy self-evolution engine module, the present invention overcomes the defect of the prior art lacking dynamic adaptability. The environmental situation awareness module collects the meteorological mutation index and pest density in real time, and can keenly capture the changes in the forest environment. Once environmental mutations are detected, such as extreme weather or pest outbreaks, the monitoring strategy self-evolution engine module will quickly respond. It will optimize the collaborative monitoring weights of drones, fixed nodes, and satellites according to the environmental situation parameters, and reasonably arrange the monitoring resources. In extreme weather, the sensor modality switching module can switch the drone to synthetic aperture radar imaging to ensure that data collection is not affected by bad weather. When pest outbreaks occur, the thermal infrared camera and vibration sensor are activated to jointly locate the core area of pests, improving the monitoring pertinence and accuracy. Through the dynamic weight formula and modality switching mechanism, the entire system realizes the automatic adjustment of the monitoring strategy, has the "quasi-living body" adaptability to handle emergencies, ensures effective forest monitoring in various complex environments, avoids the problems of data collection and model analysis failure, and significantly improves the reliability and practicality of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a schematic diagram of the system process in an embodiment of the present invention;

[0060] Figure 2 It is a schematic diagram of the method steps in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0061] The following further describes the specific embodiments of the present invention with reference to the drawings. It should be noted here that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation to the present invention.

[0062] In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0063] Please refer to the attached Figure 1 - attached Figure 2 , a forest degradation monitoring and evaluation system based on machine vision of the present invention, the system includes a multi-source data collection module, a multi-dimensional dynamic degradation index calculation module, an environmental situation awareness module, a monitoring strategy self-evolution engine module, a sensor modality switching module, an edge-cloud collaborative computing module, and a visualization decision terminal module;

[0064] Multi-source data acquisition module: Integrates an unmanned aerial vehicle (UAV) multi-spectral camera, a ground capacitive soil sensor, a meteorological API interface, and a sound recognition unit. The vegetation image data collected by the UAV multi-spectral camera is transmitted in real-time via a wireless transmission module to the UAV terminal in the multi-dimensional dynamic degradation index calculation module and the edge-cloud collaborative computing module. The soil moisture data measured by the ground capacitive soil sensor is transmitted wirelessly to the multi-dimensional dynamic degradation index calculation module. The meteorological API interface obtains meteorological data, including temperature, air pressure, rainfall, etc. On the one hand, it is transmitted to the environmental situation perception module for calculating the meteorological mutation index, and on the other hand, it is also transmitted to the multi-dimensional dynamic degradation index calculation module. The human activity sound data detected by the sound recognition unit is transmitted to the multi-dimensional dynamic degradation index calculation module for calculating the human activity density parameter, and is also transmitted to the monitoring strategy self-evolution engine module to provide a reference for formulating monitoring strategies. The multi-source data acquisition module obtains vegetation index, soil moisture, rainfall, and human activity data in real-time;

[0065] Multi-dimensional dynamic degradation index calculation module: Receives the data transmitted by the multi-source data acquisition module and calculates the dynamic degradation index based on a formula. The specific calculation formula is as follows:

[0066]

[0067] Where: MDDI represents the degradation index, λ, μ, ν, and ξ are adaptively allocated according to the forest area type through the Analytic Hierarchy Process (AHP), and λ + μ + ν + ξ = 1. The multi-dimensional dynamic degradation index calculation module transmits the calculated data to the visualization decision terminal module for generating a degradation level heat map and governance suggestions. On the other hand, it is transmitted to the LSTM network in the cloud as one of the input data for predicting the degradation trend;

[0068] When calculating If SM wet = 0, it is replaced with 1% of the historical average to avoid division by zero errors;

[0069] Environmental situation perception module: Calculates the meteorological mutation index CMI and pest density in real-time through a weather station and a bionic vibration sensor. The change data of meteorological parameters such as temperature and air pressure collected by the weather station is used to calculate the meteorological mutation index, and the calculation result is transmitted to the monitoring strategy self-evolution engine module. The 200Hz - 500Hz characteristic vibration signals detected by the bionic vibration sensor are processed and calculated to obtain pest density data, which is also transmitted to the monitoring strategy self-evolution engine module to provide environmental situation information for the dynamic generation of monitoring strategies;

[0070] Monitoring Strategy Self-Evolution Engine Module: Receives the meteorological mutation index and pest density data transmitted by the environmental situation awareness module, generates dynamic task scheduling instructions based on this data. When the meteorological mutation index exceeds the threshold or the pest density reaches a certain level, calculates and generates scheduling instructions for the UAV cluster according to a specific formula, and transmits them to the UAV terminal to control the execution of the UAV's monitoring tasks. At the same time, it will also trigger the operation of the sensor modality switching module according to the data situation;

[0071] Sensor Modality Switching Module: After receiving the trigger instruction from the monitoring strategy self-evolution engine module, enables the synthetic aperture radar in extreme weather, and the data collected by it is transmitted to the edge-cloud collaborative computing module for processing. When pests break out, it switches to a thermal infrared imager, and the collected data is also transmitted to the edge-cloud collaborative computing module for subsequent analysis and processing;

[0072] Edge-Cloud Collaborative Computing Module: Deploys a lightweight model at the UAV terminal, receives the data transmitted by the UAV multi-spectral camera, synthetic aperture radar, and thermal infrared imager, realizes real-time segmentation of vegetation coverage. On the one hand, the segmentation result is fed back to the monitoring strategy self-evolution engine module to provide a basis for further adjustment of the monitoring strategy. On the other hand, it is transmitted to the LSTM network in the cloud, receives the MDDI historical sequence from the multi-dimensional dynamic degradation index calculation module, meteorological forecast data, and the segmentation result data transmitted by the UAV terminal, predicts the degradation trend, and transmits the prediction result to the visualization decision terminal module;

[0073] Visualization Decision Terminal Module: Receives the degradation index transmitted by the multi-dimensional dynamic degradation index calculation module and the degradation trend prediction result transmitted by the edge-cloud collaborative computing module, maps the degradation index value to a heat map, and matches the governance strategy according to the degradation level and cause type (pest / deforestation / fire), and generates a multi-dimensional report including the replanting area and law enforcement priority.

[0074] In an embodiment of the present invention: In the multi-dimensional dynamic degradation index calculation module, EVI baseline represents the baseline enhanced vegetation index, which is the enhanced vegetation index value of a certain forest area in the past relatively stable state obtained by calling the historical database. After matching with the current GPS coordinates, it is used as a comparison benchmark to measure the change of the vegetation index at the current moment. The vegetation index is an important indicator to measure the growth status and coverage of vegetation. Compared with other vegetation indexes, the enhanced vegetation index can better eliminate the influence of noise such as atmosphere and soil background, and more accurately reflect the true state of vegetation. EVI current represents the enhanced vegetation index at the current moment, which is obtained by collecting the current vegetation image data by the UAV multi-spectral camera in the multi-source data acquisition module and processing and analyzing it, and is used to compare with EVI baseline, to reflect the real-time changes in vegetation growth conditions, the SM dry represents the soil moisture in the dry state, which is the soil moisture data measured by a ground capacitive soil sensor during a monthly time period when the soil moisture is continuously lower than 60% of the historical average soil moisture in this forest area. SM wet , represents the soil moisture in the wet state, which is the data obtained by measuring with a ground capacitive soil sensor during a monthly time period when the soil moisture is continuously higher than 80% of the historical average soil moisture in this forest area. RF anomaly represents the rainfall anomaly value, which is the difference obtained by comparing the current rainfall data acquired through a meteorological API interface with the average rainfall in the same historical period of this area. RF avg represents the average rainfall in the same historical period of this area, which is obtained based on long-term statistical analysis of meteorological data and serves as a reference standard for measuring whether the current rainfall is abnormal. represents the density of the human activity environment situation perception module, which is obtained by detecting the sound pattern of logging operations through a sound recognition unit and combining with the YOLOv7 model to identify fire points and vehicle targets for statistics.

[0075] In one embodiment of the present invention: The environment situation perception module includes:

[0076] The calculation formula for the meteorological mutation index CMI is as follows:

[0077]

[0078] Among them, ΔT represents the average daily temperature change rate, and ΔP represents the average daily air pressure change rate. The data source is a distributed meteorological station network;

[0079] Pest density density Detect the characteristic vibration signal of 200Hz - 500Hz through a bionic vibration sensor and calculate the number of peaks per unit time.

[0080] In one embodiment of the present invention: The monitoring strategy self-evolution engine module performs the following operations:

[0081] When the CMI exceeds the set threshold, calculate the priority weight of the UAV to execute the monitoring task according to the formula. The specific calculation formula is as follows:

[0082] W drone = α·CMI + β·Pest density

[0083] Among them, α and β are weight coefficients optimized online through reinforcement learning. According to the calculated W drone Generate a scheduling instruction for the UAV cluster, transmit it to the UAV terminal, and control the execution of the monitoring task of the UAV.

[0084] When Pest density > 5 times per minute, it triggers the collaborative positioning of the core pest area by the thermal infrared camera and the vibration sensor;

[0085] The reinforcement learning training framework for α and β is as follows:

[0086]

[0087] The training data is sourced from the environmental parameters and task execution effects in historical monitoring tasks.

[0088] In one embodiment of the present invention: In the edge-cloud collaborative computing module, a lightweight U-Net model is deployed on the drone terminal. The lightweight U-Net model is processed by model compression and quantization optimization techniques to reduce the consumption of computing resources while ensuring the segmentation accuracy. The cloud uses an LSTM network to predict the degradation trend, and the input data includes the MDDI historical sequence and weather forecasts;

[0089] The lightweight U-Net model reduces the number of parameters by 50% through channel pruning and reduces the computing precision by using 8-bit quantization. It is deployed under the TensorRT framework, and the inference speed is increased by 3 times.

[0090] In one embodiment of the present invention: In the visual decision terminal module, the visual decision terminal module maps the degradation index value into a three-color heat map of red (severe), orange (moderate), and green (mild). The mapping process uses a threshold division method, and according to the degradation level and cause type (pest / deforestation / fire), it matches the governance strategy and generates a multi-dimensional report including the replanting area and law enforcement priority. The report generation process combines historical governance data and expert experience;

[0091] Basis for dividing the degradation level:

[0092] Green (mild): MDDI < 0.2;

[0093] Orange (moderate): 0.2 ≤ MDDI < 0.5;

[0094] Red (severe): MDDI ≥ 0.5;

[0095] The threshold is determined by clustering analysis of historical degradation data and supports manual calibration.

[0096] In one embodiment of the present invention: After the monitoring strategy self-evolution engine module calculates W drone it will generate corresponding scheduling instructions according to different threshold ranges:

[0097] W drone≥8 (High priority): Immediately deploy all available drones, requiring them to fly to the target area at the fastest speed and along the shortest path, collect images with the highest resolution, and transmit monitoring data in real-time every 5 minutes to conduct comprehensive and high-frequency monitoring of the forest environment without blind spots;

[0098] 5 ≤ W drone <8 (Medium priority): Arrange two-thirds of the drone swarm to conduct detailed monitoring of key areas according to the pre-planned zoning. These drones need to maintain medium resolution for image collection and transmit data every 15 minutes to promptly grasp the dynamic changes in the forest;

[0099] 2 ≤ W drone <5 (Low priority): Dispatch one-fourth of the drones to perform routine patrol tasks, fly at normal speed along the default path, use low resolution for image collection, and transmit data every 30 minutes to maintain basic monitoring of the forest condition;

[0100] W drone <2 (Very low priority): Only send a few drones for general overview patrols. The flight speed is slightly slower, the image collection resolution is relatively low, and data is transmitted once an hour, mainly to confirm whether there are any abnormal situations in the overall forest.

[0101] In one embodiment of the present invention: EVI current is obtained through the following processing and analysis process: The current vegetation image data collected by the drone multi-spectral camera is first radiometrically calibrated to convert the digital quantization value obtained by the camera into an absolute radiance value, eliminating the radiation error caused by the difference in camera response and ensuring the comparability of data collected at different times and under different conditions. Subsequently, using the atmospheric correction algorithm, the effects of atmospheric molecular scattering, absorption, and aerosol scattering on light are removed to restore the true reflectance information of the ground objects. Then, image mosaicking and cropping techniques are used to stitch multiple collected images into a complete large-area image and crop it according to the boundary of the research area to obtain the target area image. After that, the supervised classification algorithm is used to classify the pixels in the image into vegetation and non-vegetation categories based on the known spectral characteristics of vegetation samples, and the vegetation area is extracted. Finally, within the vegetation area, EVI is calculated according to the formula current EVI current The specific calculation formula is as follows:

[0102]

[0103] where ρ NIR ρ RED and ρ Blue respectively represent the reflectance of the near-infrared band, red band, and blue band.

[0104] In one embodiment of the present invention: A method for monitoring and evaluating forest degradation based on machine vision, the method comprising the following steps:

[0105] Data acquisition step: Using a multi-source data acquisition module integrated with a drone multispectral camera, a ground capacitive soil sensor, a meteorological API interface, and a sound recognition unit to collect data. Among them, the drone multispectral camera collects vegetation image data and transmits it in real time via a wireless transmission module to the drone terminal in the multi-dimensional dynamic degradation index calculation module and the edge-cloud collaborative calculation module. The ground capacitive soil sensor measures soil humidity data and transmits it wirelessly to the multi-dimensional dynamic degradation index calculation module. The meteorological API interface obtains meteorological data and transmits it to the environmental situation awareness module for calculating the meteorological mutation index and the multi-dimensional dynamic degradation index calculation module respectively. The sound recognition unit detects human activity sound data and transmits it to the multi-dimensional dynamic degradation index calculation module for calculating the human activity density parameter, and at the same time transmits it to the monitoring strategy self-evolution engine module to realize the real-time acquisition of vegetation index, soil humidity, rainfall, and human activity data;

[0106] Data calculation and analysis step: The multi-dimensional dynamic degradation index calculation module receives the collected data and calculates the degradation index based on the formula The environmental situation awareness module collects meteorological parameter change data through a meteorological station and calculates the meteorological mutation index using the formula At the same time, it detects the 200Hz - 500Hz characteristic vibration signal through a bionic vibration sensor to calculate the pest density, and transmits the meteorological mutation index and pest density data to the monitoring strategy self-evolution engine module;

[0107] Monitoring strategy formulation and execution step: The monitoring strategy self-evolution engine module receives the data transmitted from the environmental situation awareness module. When the meteorological mutation index exceeds the threshold or the pest density reaches a certain level, it calculates the priority weight of the drone to execute the monitoring task according to the formula W drone = α·CMI + β·Pest density Generates a scheduling instruction for the drone cluster according to the calculation result, transmits it to the drone terminal to control the execution of the monitoring task, and at the same time triggers the sensor mode switching module to work according to the data situation. In extreme weather, the sensor mode switching module enables the synthetic aperture radar, and switches to the thermal infrared imager during pest outbreaks. The collected data are all transmitted to the edge-cloud collaborative calculation module. The drone terminal of the edge-cloud collaborative calculation module uses the lightweight U-Net model to perform real-time segmentation of the received data for vegetation coverage. On the one hand, the segmentation result is fed back to the monitoring strategy self-evolution engine module, and on the other hand, it is transmitted to the cloud LSTM network. The cloud LSTM network combines the MDDI historical sequence of the multi-dimensional dynamic degradation index calculation module and the meteorological forecast data to predict the degradation trend;

[0108] Result presentation and decision-making steps: The visual decision-making terminal module receives the degradation index from the multi-dimensional dynamic degradation index calculation module and the degradation trend prediction result from the edge-cloud collaborative calculation module, maps the degradation index value into a three-color heat map of red (severe), orange (moderate), and green (mild) through the threshold division method, matches the governance strategy according to the degradation level and cause type (pest / deforestation / fire), combines historical governance data and expert experience, generates a multi-dimensional report including the replanting area and law enforcement priority, and provides a decision-making basis for forest resource management.

[0109] Example 1. Please refer to the appendix Figure 1 - Appendix Figure 2 , select the Southeast Asian tropical rainforest area as the monitoring object. The area of this region is about 100 square kilometers, with rich vegetation species and a complex ecosystem;

[0110] In the data collection step, the multi-source data collection module starts to work. The drone multi-spectral camera flies according to the preset route, collects a vegetation image data every 50 meters, and transmits it to the drone terminal in the multi-dimensional dynamic degradation index calculation module and the edge-cloud collaborative calculation module in real time through the wireless transmission module. The ground capacitive soil sensors are evenly distributed with 50 monitoring points in this area, measure the soil humidity data once an hour, and transmit it to the multi-dimensional dynamic degradation index calculation module wirelessly. The meteorological API interface continuously obtains meteorological data such as temperature, air pressure, and rainfall in this area. Among them, the temperature data is updated every 10 minutes, the air pressure data is updated every 15 minutes, and the rainfall data is updated in real time. On the one hand, these meteorological data are transmitted to the environmental situation perception module for calculating the meteorological mutation index, and on the other hand, they are transmitted to the multi-dimensional dynamic degradation index calculation module. The sound recognition unit arranges 10 monitoring points in this area, detects the human activity sound data in real time, transmits it to the multi-dimensional dynamic degradation index calculation module for calculating the human activity density parameter, and at the same time transmits it to the monitoring strategy self-evolution engine module;

[0111] In the data calculation and analysis step, the multi-dimensional dynamic degradation index calculation module receives the collected data. At a certain moment, the baseline enhanced vegetation index EVI of this area is obtained by calling the historical database baseline is 0.8. The current vegetation image data collected by the drone multi-spectral camera, after a series of processes such as radiometric calibration, atmospheric correction, image mosaicking and cropping, and supervised classification, calculates the current enhanced vegetation index as EVI current 0.75. In the past month, through the measurement of the ground capacitive soil sensor, the soil humidity SM in the dry state is determined dryis 0.2 (the average value measured during the period when the soil humidity in the month is continuously lower than 60% of the historical average soil humidity), and the soil humidity in the wet state is SM wet 0.5 (the average value measured during the period when the soil humidity in the month is continuously higher than 80% of the historical average soil humidity), the current rainfall obtained from the meteorological API interface is 50 mm, and the average rainfall in the same period of this area in history is RF avg is 40 mm, so the rainfall anomaly value is RF anomaly is 10 mm. The sound recognition unit combines the YOLOv7 model to count the frequency of human activities per unit area is 3 times per square kilometer (the main detected human activities are the entry of a small number of scientific research personnel and occasional signs of illegal logging). According to the analytic hierarchy process (AHP), for the tropical rainforest area, the weight coefficients are determined as λ = 0.4, μ = 0.3, v = 0.2, ξ = 0.1. Through the multi-dimensional dynamic degradation index calculation formula:

[0112]

[0113] Calculated

[0114] The environmental situation awareness module collects meteorological parameter change data through a weather station. On a certain day, the average daily cycle temperature change rate ΔT is 2 °C / day, and the average daily cycle air pressure change rate ΔP is 500 Pa / day. According to the meteorological mutation index calculation formula (where Δt is 1 day), calculated The bionic vibration sensor detects 200 Hz - 500 Hz characteristic vibration signals, and the number of pest characteristic vibration peaks Pest detected within 1 minute density is 3 times, the sampling period is 1 minute, and the pest density is calculated to be 3 times /

[0115] minute, and the meteorological mutation index and pest density data are transmitted to the monitoring strategy self-evolution engine module;

[0116] Supervised classification uses the support vector machine (SVM). The training set contains 1000 vegetation samples and 500 non-vegetation samples, and the features are the reflectance of the near-infrared, red, and blue light bands and the NDVI index.

[0117] The monitoring strategy self-evolution engine module receives the data transmitted from the environmental situation awareness module. Since CMI = 7 exceeds the set threshold (the threshold is 5), according to the formula W drone = α·CMI + β·Pest density Calculate the priority weight of the drone to perform the monitoring task (α = 0.6, β = 0.4 obtained through online optimization by reinforcement learning), then W drone= α·CMI + β·Pest density = 0.6

[0118] ×7 + 0.4×3 = 5.4. According to the calculation result, it belongs to a medium-priority task. Generate scheduling instructions for the UAV cluster, arrange two-thirds of the UAV cluster, and conduct detailed monitoring of key areas according to the pre-planned partition. These UAVs maintain medium-resolution image acquisition and transmit data every 15 minutes. At the same time, since Pest density = 3 times / minute is less than 5 times / minute, the collaborative positioning of the thermal infrared camera and the vibration sensor for the core area of pests is not triggered for the time being.

[0119] The UAV terminal of the edge-cloud collaborative computing module uses the lightweight U-Net model to perform real-time segmentation of the received data for vegetation coverage. After segmentation, it is determined that the vegetation coverage area of this area is 70 square kilometers, which is 2 square kilometers less than the previous monitoring period. On the one hand, the segmentation result is fed back to the monitoring strategy self-evolution engine module, and on the other hand, it is transmitted to the cloud LSTM network. The cloud LSTM network combines the MDDI historical sequence of the multi-dimensional dynamic degradation index calculation module and meteorological forecast data to predict the degradation trend. After prediction by the trained LSTM network, the MDDI of this area may rise to 0.25 in the next month, indicating that the forest degradation has an aggravating trend.

[0120] The visualization decision terminal module receives the degradation index MDDI = 0.21 of the multi-dimensional dynamic degradation index calculation module and the degradation trend prediction result of the edge-cloud collaborative computing module, maps the degradation index value into a heat map through the threshold division method. Since MDDI = 0.21, it is determined as mild degradation and mapped to green. According to the degradation level and cause type (initially judged as possibly existing slight illegal logging and natural vegetation growth changes), combined with historical governance data and expert experience, match the governance strategy. The generated multi-dimensional report recommends: strengthening the daily patrol and law enforcement of this area, focusing on areas with frequent human activities; marking the areas with reduced vegetation coverage and planning to replant in the next rainy season, with a replanting area of about 5000 square meters.

[0121] Example 2. Please refer to the appendix Figure 1 -appendix Figure 2 , monitoring of arid forest areas

[0122] Select an arid forest area with an area of about 80 square kilometers. The vegetation in this area is relatively sparse and the ecological system is relatively fragile.

[0123] During the data collection process, the drone multi-spectral camera collects images at a relatively high flight altitude and with a large shooting interval. One image is collected every 100 meters to ensure coverage of the entire forest area. 30 monitoring points are arranged for the ground capacitive soil sensors. Since the soil humidity change in arid areas is relatively small, the soil humidity data is measured every 2 hours. Meteorological data is obtained through the meteorological API interface. The temperature data is updated every 15 minutes, the air pressure data is updated every 20 minutes, and the rainfall data is updated in real-time. 8 monitoring points are arranged for the sound recognition unit to detect human activity sounds.

[0124] In the data calculation and analysis stage, the multi-dimensional dynamic degradation index calculation module receives data, and the obtained EVI baseline is 0.5, and the processed EVI current is 0.45. Within a statistically counted month, SM dry is 0.1 (the average value measured during the period when the soil humidity is continuously lower than 60% of the historical average soil humidity within this month), and since there are fewer wet periods in the arid forest area, SM wet is 0.3 (the average value measured during the period when the soil humidity is continuously higher than 80% of the historical average soil humidity within this month). The current rainfall is 10 mm, RF avg is 15 mm, RF anomaly is -5 mm. Through sound recognition and image detection, is 1 time / square kilometer. For the arid forest area, the weight coefficients λ = 0.2, μ = 0.4, v = 0.3, ξ = 0.1 are determined through AHP, and MDDI is calculated:

[0125]

[0126] In the environmental situation awareness module, the ΔT collected by the weather station is 3 °C / day, and ΔP is 400 Pa / day. Calculate The bionic vibration sensor detects 1 time within 1 minute, Δt is 1 minute, and Pest density = 1 time / minute.

[0127] The monitoring strategy self-evolution engine module receives data and calculates W drone (α = 0.7, β = 0.3), W drone = α·CMI + β·Pest density = 0.7×7 + 0.3×1 = 5.2, which belongs to a medium-priority task. Arrange the corresponding drone cluster to monitor the key area and transmit data back every 15 minutes.

[0128] The edge-cloud collaborative computing module performs vegetation coverage segmentation and finds that the vegetation coverage area is 40 square kilometers, showing no obvious change compared with the previous cycle. The cloud LSTM network predicts that the MDDI may remain at about 0.05 in the next month.

[0129] The visual decision-making terminal module determines it as mild degradation according to MDDI = 0.03, generates a report in combination with the causes (mainly the impact of climate drought on vegetation growth), suggests strengthening the construction of irrigation facilities in the forest area, plans to lay drip irrigation systems in some areas, with a coverage area of about 10 square kilometers, and continuously monitors the vegetation growth status.

[0130] According to the content of the above embodiments, by integrating multiple functional modules such as a multi-source data acquisition module, a multi-dimensional dynamic degradation index calculation module, an environmental situation perception module, and a monitoring strategy self-evolution engine module, the deficiencies in the prior art in forest degradation monitoring and assessment are effectively solved, the efficient monitoring and precise assessment and governance of forest resources are realized, and it has significant advantages and broad application prospects in the field of forest resource management.

[0131] In terms of assessment capabilities, the multi-source data acquisition module collects data on vegetation, soil, climate, and human activities in real time, providing a rich and comprehensive information basis for assessment. The multi-dimensional dynamic degradation index calculation module integrates this data, calculates the degradation index using a unique formula, comprehensively considers the impact of various factors on forest degradation, breaks through the limitations of the prior art that rely on a single indicator or a static weight model, and the weights are automatically assigned according to the forest area type, making the assessment results more in line with the actual situation of different forest areas. The visual decision-making terminal module maps the degradation index value to a heat map based on this data and calculation results, and generates a multi-dimensional report including replanting area, law enforcement priority, etc., providing a scientific basis for forest resource management, realizing precise assessment and differential governance, and greatly improving the comprehensiveness and accuracy of forest degradation assessment.

[0132] In terms of dynamic adaptability, the environmental situation perception module collects the meteorological mutation index and pest density in real time, and keenly captures changes in the forest environment. Once environmental mutation situations such as extreme weather or pest outbreaks are detected, the monitoring strategy self-evolution engine module quickly responds, optimizes the collaborative monitoring weights of drones, fixed nodes, and satellites according to the environmental situation parameters, and reasonably allocates monitoring resources. In extreme weather, the sensor modality switching module enables the drone to switch to synthetic aperture radar imaging to ensure that data collection is not interfered by bad weather. In the event of a pest outbreak, a thermal infrared camera and a vibration sensor are activated to jointly locate the core area of the pest, improving the monitoring pertinence and accuracy. The entire system realizes the automatic adjustment of the monitoring strategy through a dynamic weight formula and a modality switching mechanism, has the "quasi-living body" adaptability to respond to emergencies, effectively avoids the problems of data collection and model analysis failure, and significantly improves the reliability and practicality of the system in complex environments.

[0133] In summary, through the collaborative cooperation among various modules, the present invention demonstrates powerful functions in forest degradation monitoring and assessment, provides strong technical support for the sustainable management of forest resources, and has broad application prospects in the forestry field.

[0134] Although the present invention is disclosed above in a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of the present invention. Therefore, any modifications, equivalent changes, and decorations made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention shall fall within the protection scope defined by the claims of the present invention.

Claims

1. A forest degradation monitoring and evaluation system based on machine vision, characterized in that: The system includes a multi-source data acquisition module, a multi-dimensional dynamic degradation index calculation module, an environmental situation awareness module, a monitoring strategy self-evolution engine module, a sensor modality switching module, an edge-cloud collaborative computing module, and a visualization decision terminal module; Multi-source data acquisition module: Integrates an unmanned aerial vehicle (UAV) multi-spectral camera, a ground capacitive soil sensor, a meteorological API interface, and a sound recognition unit; Multi-dimensional dynamic degradation index calculation module: Receives the data transmitted by the multi-source data acquisition module and dynamically calculates the degradation index based on a formula. The specific calculation formula is as follows: Where: MDDI represents the degradation index, and λ, μ, ν, and ξ are adaptively allocated according to the forest area type through the Analytic Hierarchy Process (AHP), and λ + μ + ν + ξ = 1. The multi-dimensional dynamic degradation index calculation module transmits the calculated data to the visualization decision terminal module and, on the other hand, transmits it to the LSTM network in the cloud; Environmental situation awareness module: Real-time calculates the meteorological mutation index CMI and pest density through a weather station and a bionic vibration sensor; Monitoring strategy self-evolution engine module: Receives the meteorological mutation index and pest density data transmitted by the environmental situation awareness module and generates dynamic task scheduling instructions based on these data; Sensor modality switching module: After receiving the trigger instruction from the monitoring strategy self-evolution engine module, enables a synthetic aperture radar in extreme weather, and the data collected by it is transmitted to the edge-cloud collaborative computing module for processing. When a pest outbreak occurs, it switches to a thermal infrared imager, and the collected data is also transmitted to the edge-cloud collaborative computing module for subsequent analysis and processing; Edge-cloud collaborative computing module: Deploys a lightweight model at the UAV terminal, receives the MDDI historical sequence, meteorological forecast data from the multi-dimensional dynamic degradation index calculation module, and the segmentation result data transmitted by the UAV terminal, predicts the degradation trend, and transmits the prediction result to the visualization decision terminal module; Visualization decision terminal module: Outputs a heat map of the degradation level and governance suggestions.

2. The forest degradation monitoring and evaluation system based on machine vision according to claim 1, characterized in that: In the multi-dimensional dynamic degradation index calculation module, the EVI baseline represents the baseline enhanced vegetation index, which is the enhanced vegetation index value of a certain forest area in the past relatively stable state obtained by calling the historical database. After matching with the current GPS coordinates, it is used as a comparison benchmark to measure the change of the vegetation index at the current moment. The vegetation index is an important indicator to measure the vegetation growth status, coverage, etc. Compared with other vegetation indexes, the enhanced vegetation index can better eliminate the influence of noise such as atmosphere and soil background, and more accurately reflect the true state of the vegetation. The EVI current represents the enhanced vegetation index at the current moment, which is obtained by collecting the current vegetation image data by the drone multispectral camera in the multi-source data acquisition module and then processed and analyzed, and is used to compare with the EVI baseline to reflect the real-time change of the vegetation growth status. The SM dry represents the soil moisture in the dry state, which is the soil moisture data measured by the ground capacitive soil sensor during the time period with a monthly time cycle and the soil moisture continuously lower than 60% of the historical average soil moisture in this forest area during this period. The SM wet represents the soil moisture in the wet state, which is the data obtained by measuring the ground capacitive soil sensor during the time period with a monthly time cycle and the soil moisture continuously higher than 80% of the historical average soil moisture in this forest area during this period. The RF anomaly represents the rainfall anomaly value, which is the difference obtained by comparing the current rainfall data obtained through the meteorological API interface with the average rainfall in the same historical period of this area. The RF avg represents the average rainfall in the same historical period of this area. The represents the density of the human activity environment situation perception module, which is obtained by detecting the logging operation sound pattern through the sound recognition unit and then statistically analyzing by combining the YOLOv7 model to identify fire points and vehicle targets.

3. The forest degradation monitoring and evaluation system based on machine vision according to claim 1, characterized in that The environmental situation awareness module includes: The calculation formula for calculating the meteorological mutation index CMI is as follows: Where, ΔT represents the average daily cycle temperature change rate, and ΔP represents the average daily cycle air pressure change rate; The pest density Pest density Detect the characteristic vibration signal of 200Hz - 500Hz through a bionic vibration sensor and calculate the number of peaks per unit time.

4. A machine vision-based forest degradation monitoring and evaluation system according to claim 1, characterized in that: The monitoring strategy self-evolution engine module performs the following operations: When the CMI exceeds the set threshold, calculate the priority weight of the UAV to perform the monitoring task according to the formula. The specific calculation formula is as follows: W drone = α · CMI + β · Pest density where α and β are weight coefficients optimized online through reinforcement learning, and according to the calculated W drone generate scheduling instructions for the UAV swarm, transmit them to the UAV terminals, and control the execution of the monitoring tasks of the UAVs; When Pest density > 5 times per minute, trigger the collaborative positioning of the core pest area by the thermal infrared camera and the vibration sensor.

5. The forest degradation monitoring and evaluation system based on machine vision according to claim 1, characterized in that: In the edge-cloud collaborative computing module, a lightweight U-Net model is deployed at the UAV terminal. The lightweight U-Net model is processed by model compression and quantization optimization techniques to reduce the consumption of computing resources while ensuring the segmentation accuracy. The cloud uses an LSTM network to predict the degradation trend, and the input data includes the MDDI historical sequence and meteorological forecast.

6. A forest degradation monitoring and evaluation system based on machine vision according to claim 1, characterized in that: In the visualization decision terminal module, the visualization decision terminal module maps the degradation index value into a three-color heat map of red (severe), orange (moderate), and green (mild). The mapping process uses a threshold division method. According to the degradation level and cause type (pest damage / felling / fire), it matches the treatment strategies and generates a multi-dimensional report including the replanting area and law enforcement priority. The report generation process combines historical treatment data and expert experience.

7. The forest degradation monitoring and evaluation system based on machine vision according to claim 4, characterized in that: After the monitoring strategy self-evolution engine module calculates W drone it will generate corresponding scheduling instructions according to different threshold ranges: W drone ≥8 (High priority): Immediately deploy all available drones, requiring them to fly to the target area at the fastest speed and along the shortest path, collect images with the highest resolution, and transmit monitoring data in real time every 5 minutes to conduct comprehensive and high-frequency monitoring of the forest environment without blind spots; 5≤W drone <8 (Medium priority): Arrange two-thirds of the UAV swarm to conduct detailed monitoring of key areas according to the pre-planned partitions. These UAVs need to maintain medium-resolution image acquisition and transmit data every 15 minutes; 2 ≤ W drone < 5 (low priority): Dispatch one - quarter of the drones to perform routine patrol tasks, fly at normal speed along the default path, collect images with low resolution, and transmit data every 30 minutes; W drone <2 (Very low priority): Only send out individual drones for general inspection, with a slightly slower flight speed, lower image resolution for collection, and data transmission once per hour.

8. The forest degradation monitoring and evaluation system based on machine vision according to claim 2, wherein: The EVI current is obtained through the following processing and analysis process: For the current vegetation image data collected by the drone multispectral camera, first, radiometric calibration is performed to convert the digital quantization value obtained by the camera into an absolute radiance value, eliminating the radiometric error caused by the difference in camera response and ensuring the comparability of data collected at different times and under different conditions. Subsequently, using the atmospheric correction algorithm, the effects of atmospheric molecular scattering, absorption, and aerosol scattering on light are removed to restore the true reflectance information of the ground objects. Then, image mosaicking and cropping techniques are adopted to stitch multiple collected images into a complete large-area image and crop it according to the boundary of the research area to obtain the target area image. After that, using the supervised classification algorithm, based on the known spectral characteristics of vegetation samples, the pixels in the image are classified into vegetation and non-vegetation categories, and the vegetation area is extracted. Finally, within the vegetation area, the EVI is calculated according to the formula current , the EVI current The specific calculation formula is as follows: Among them, ρ NIR , ρ RED and ρ Blue respectively represent the reflectance in the near-infrared band, red band and blue band.

9. A method applicable to the machine vision-based forest degradation monitoring and assessment system according to any one of claims 1-8, characterized in that, The method includes the following steps: Data collection step: Use a multi-source data collection module integrated with a drone multi-spectral camera, a ground capacitive soil sensor, a meteorological API interface, and a sound recognition unit to collect data. Among them, the drone multi-spectral camera collects vegetation image data and sends it to the drone terminal in the multi-dimensional dynamic degradation index calculation module and the edge-cloud collaborative calculation module in real time through a wireless transmission module. The ground capacitive soil sensor measures the soil humidity data and transmits it to the multi-dimensional dynamic degradation index calculation module wirelessly. The meteorological API interface obtains meteorological data and transmits it to the environmental situation awareness module for calculating the meteorological mutation index and the multi-dimensional dynamic degradation index calculation module respectively. The sound recognition unit detects human activity sound data and transmits it to the multi-dimensional dynamic degradation index calculation module for calculating the human activity density parameter, and at the same time transmits it to the monitoring strategy self-evolution engine module to realize the real-time acquisition of vegetation index, soil humidity, rainfall, and human activity data. Data calculation and analysis steps: The multi-dimensional dynamic degradation index calculation module receives the collected data and calculates the degradation index based on the formula The environmental situation awareness module collects the meteorological parameter change data through a weather station and calculates the meteorological mutation index using the formula At the same time, it detects the characteristic vibration signals of 200Hz - 500Hz through a bionic vibration sensor to calculate the pest density, and transmits the meteorological mutation index and pest density data to the monitoring strategy self-evolution engine module; Monitoring strategy formulation and execution steps: The monitoring strategy receives data transmitted from the environmental situation awareness module by the self-evolving engine module of the monitoring strategy. When the meteorological mutation index exceeds the threshold or the pest density reaches a certain level, according to the formula W drone = α·CMI + β·Pest density Calculate the priority weight of the UAV to execute the monitoring task. Generate a scheduling instruction for the UAV cluster according to the calculation result, and transmit it to the UAV terminal to control the execution of the monitoring task. At the same time, trigger the sensor modality switching module to work according to the data situation. In extreme weather, the sensor modality switching module enables the synthetic aperture radar, and switches to the thermal infrared imager during pest outbreaks. The collected data are all transmitted to the edge-cloud collaborative computing module. The UAV terminal of the edge-cloud collaborative computing module uses the lightweight U-Net model to perform real-time segmentation of the vegetation coverage on the received data. The segmentation result is on the one hand fed back to the self-evolving engine module of the monitoring strategy, and on the other hand transmitted to the cloud LSTM network. The cloud LSTM network combines the MDDI historical sequence of the multi-dimensional dynamic degradation index calculation module and the meteorological forecast data to predict the degradation trend; Result presentation and decision-making step: The visualization decision terminal module receives the degradation index from the multi-dimensional dynamic degradation index calculation module and the degradation trend prediction result from the edge-cloud collaborative calculation module, maps the degradation index value into a three-color heat map of red (severe), orange (moderate), and green (mild) through a threshold division method. According to the degradation level and cause type (pest damage / felling / fire), combined with historical treatment data and expert experience, it matches the treatment strategies and generates a multi-dimensional report including the replanting area and law enforcement priority, providing a decision-making basis for forest resource management.

Images