A method and system for unmanned aerial vehicle forest damage monitoring based on time-aware attention

Abstract

This invention belongs to the field of data processing technology and proposes a method and system for monitoring forest pests using unmanned aerial vehicles (UAVs) based on time-aware attention. Specifically, the UAV collects images to construct a forest pest image sequence library. Using this library, a pest perception model based on time-aware attention is constructed to obtain pest-related vectors. A K-layer perceptron is then built based on these vectors. Image descriptors are extracted from the K-layer perceptron to obtain pest monitoring scores. Pest images are collected and pest areas are marked using these scores. Pest monitoring scores are continuously acquired for each pest area to calculate the number of forest pest occurrence trajectories. Finally, the forest pest monitoring results are determined based on the number of occurrence trajectories. This method achieves micro-scale modeling of insect population density fluctuations and the lag effect of control measures. It utilizes time-series awareness to enhance the hierarchical memory of historical intervention responses and inflection point sensitivity, thereby reducing power consumption and computing power, saving storage resources, improving endurance, and preventing signal anxiety in forest scenarios.

Description

Technical Field

[0001] This invention belongs to the field of forest damage identification and drone patrol technology, specifically relating to a drone forest damage monitoring method and system based on time-aware attention. Background Technology

[0002] With the maturation of drone technology, drones are commonly used as health monitoring devices for forests. Their high degree of freedom, flexibility, and low-cost deployment significantly improve the efficiency of forest pest monitoring. However, current drone technology for monitoring forest pests faces significant bottlenecks. Firstly, there are geographical limitations. Common drone methods for pest monitoring rely on image transmission. For example, the forest pest monitoring drone with publication number CN114313260B requires transmitting image data captured by a camera back to the terminal control equipment to determine if trees in the forest are affected by pests. Therefore, such drones are still constrained by their signal coverage range, which presents the common challenge of large forest areas with limited signal equipment. Secondly, there are challenges related to drone endurance, including battery and storage life. These lightweight resources are crucial for effectively releasing the monitoring function in drone-based forest pest monitoring. Directly collecting and storing image data usually requires sacrificing image clarity or battery life to ensure monitoring coverage in a specific area. Both lower-resolution image data and more frequent monitoring tasks significantly reduce the efficiency of drone-based forest pest monitoring. Thirdly, the collected image data is usually single-point data without validity assessment, resulting in a large amount of redundant data during the drone monitoring task, and low utilization of storage and computer resources. Fourthly, current drone-based forest damage monitoring methods are based on forest damage detection without accompanying prevention and control backtesting. This means that the transformation trajectory of pests during the prevention and control process still relies on graphical comparisons between time points, ignoring the evolutionary process and making the monitoring function always lack comprehensive prevention and control monitoring capabilities. Therefore, there is an urgent need for a drone-based forest damage monitoring method and system based on time-aware attention. Summary of the Invention

[0003] The purpose of this invention is to propose a method and system for monitoring forest damage by drones based on time-aware attention, so as to solve one or more technical problems existing in the prior art, and at least provide a beneficial option or create conditions.

[0004] To achieve the above objectives, according to one aspect of the present invention, a method for monitoring forest damage using unmanned aerial vehicles (UAVs) based on time-aware attention is provided, the method comprising the following steps:

[0005] S100, using drones to collect images and construct a forest damage image sequence library;

[0006] S200, using a forest damage image sequence library to construct a time-aware attention-based pest perception model to obtain pest-related vectors;

[0007] S300, constructs a K-layer perceptron based on pest-related vectors;

[0008] S400 extracts image descriptors from the K-layer perceptron and obtains pest monitoring accuracy;

[0009] S500 collects pest images and marks pest-infested areas through pest monitoring;

[0010] S501, Calculate the number of forest pest occurrence trajectories by continuously acquiring pest monitoring data in pest-affected areas;

[0011] S502, determine the forest damage monitoring results based on the number of forest damage occurrence trajectories.

[0012] Furthermore, in step S100, the method for constructing a forest pest image sequence library by collecting images with a drone is as follows: the flight route and monitoring cycle are set according to the known locations of pests in the forest area, and images are collected by the drone at regular intervals every day to form image sequence instances; the sequence formed by arranging the images of various pests at the same location in chronological order constitutes an image sequence instance, and the forest pest image sequence library is composed of several pest image sequence instances.

[0013] The forest pest image sequence library corresponds to only one type of forest pest. The forest pest must have obvious visual symptoms, a clear development stage, and strong regional concentration. The types of forest pests can include pine sawyer beetle, gypsy moth, poplar inchworm, or pine caterpillar, etc.

[0014] The image sequence instance includes a start frame, an outbreak frame, and a prevention end frame. The start frame represents the time point in the image sequence where pest symptoms or suspected symptoms first appear. The outbreak frame is the time point in the image sequence where pest symptoms spread rapidly and reach a significant level. The prevention end frame is the time point in the image sequence where pest symptoms have been basically alleviated after human intervention, including spraying, using insect nets, or forest land management.

[0015] In the drone's flight path planning, the altitude is set to 30-60m, with a default value of 30m; the ground sampling interval is 3-5cm / pixel, with a default value of 5cm / pixel; the camera's field of view (FOV) is 84° horizontally and 63° vertically; the forward overlap rate is greater than or equal to 70%, with a default value of 70%; the lateral overlap rate is greater than or equal to 60%, with a default value of 60%; and the path mode is a zig-zag serpentine flight path.

[0016] The purpose of this step is to provide a database of forest pest image sequences with clearly defined pest types and accurate and coherent temporal and spatial characteristics for the training and deployment of subsequent time-aware pest perception models. This database serves as the basic input for training time-aware deep learning models, enabling the models to capture the temporal image evolution patterns of various pests at different developmental stages.

[0017] Further, in step S200, the method for constructing a time-aware attention-based pest perception model using a forest damage image sequence library to obtain pest-related vectors is as follows: After preprocessing the forest damage image sequence library, a deep learning model containing a time-aware attention mechanism is used to train the image sequences to obtain the pest perception model. The time-aware attention mechanism employs any one of the TimeSformer operator, ConvLSTM operator, or Temporal Attention Network operator. The output of the intermediate embedding layer of the pest perception model is extracted as a pest-related vector, the mathematical expression of which is: , where d is the feature dimension, with a value of 128 or 256.

[0018] Preprocessing is required for any image sequence instance, including image alignment, uniform size adjustment, and color normalization.

[0019] The time-aware deep learning model uses the TimeSformer operator, ConvLSTM operator, or Temporal Attention Network operator. The default operator is TimeSformer, which is a common operator specifically designed for processing video sequences or multi-temporal image sequences. Since the development of pests is usually not linear and stable but has abrupt change nodes, the TimeSformer operator has a long-distance attention mechanism in the time dimension, which is beneficial for capturing non-continuous but causally related pest dynamics across time.

[0020] The output pest perception model is a trained pest perception model, whose prediction features are occurrence confidence and control confidence. Occurrence confidence refers to the percentile value of an image at a given time point between the start frame and the outbreak frame, and control confidence refers to the percentile value of an image at a given time point between the outbreak frame and the prevention end frame. The control confidence of each image between the start frame and the outbreak frame is 0, and the occurrence confidence of each image between the outbreak frame and the prevention end frame is 1.

[0021] During the training of the pest perception model, the loss function includes cross-entropy loss, temporal consistency regularization term and attention sparsity penalty term; the optimizer adopts Adam operator, the learning rate is set to 1e-4, the number of training rounds is 100, and the data in each round is divided into 70% of the forest pest image sequence library for training, 15% for validation, and 15% for test.

[0022] Further, in step S300, the method for constructing a K-layer perceptron based on the pest-related vector is as follows: the pest-related vector is used as input to construct a feedforward neural network model containing at least K fully connected layers as a K-layer perceptron, wherein each hidden layer adopts the ReLU activation function, and the output layer outputs including occurrence confidence and control confidence, which are used to represent the temporal position state of the input image in the pest development process.

[0023] In the structural design of the K-layer perceptron, the input layer dimension is consistent with the dimension of the pest-related vector, the number of hidden layers is K-2, K≥3, with a default value of 5, each layer adopts a fully connected structure and uses the ReLU activation function; the output layer dimension is 2, corresponding to the occurrence confidence and prevention confidence; the loss function set by the K-layer perceptron adopts the mean squared error loss, which supervises the occurrence confidence and prevention confidence respectively, with a learning rate of 1e-4, and is optimized by the Adam operator, and the generalization ability is enhanced by the Dropout and BatchNorm operators.

[0024] K-layer perceptrons, as lightweight convolutional modules, can be efficiently applied to drone operating scenarios where power consumption and computing power are constrained. K-layer perceptrons can achieve lightweight and fast judgment on the drone side, which can not only free up power consumption and computing power, but also save storage resources and improve battery life. It can also prevent signal anxiety in forest scenarios, and can work normally even in forest scenarios with poor communication conditions or no signal towers.

[0025] Further, in step S400, the method for extracting image descriptors from the K-layer perceptron and obtaining pest monitoring degree is as follows: the UAV inputs the collected images into the K-layer perceptron to obtain image descriptors and obtains the occurrence confidence and control confidence of real-time monitoring through nonlinear mapping processing, which are defined as occurrence monitoring degree and control monitoring degree, respectively, and both are classified as pest monitoring degree.

[0026] The resulting image descriptors have the same format as the pest-related vectors. As intermediate vectors within the perceptron, their mathematical expression is: , where d is the feature dimension, with a value of 128 or 256; non-linear processing uses the ReLU activation function by default.

[0027] By unifying descriptor extraction and confidence rate calculation within the K-layer perceptron, the structure of edge deployment models is simplified, making it more suitable for real-time operation scenarios of drones. The construction of the K-layer perceptron ensures that the model can operate independently as a whole, making it more friendly to subsequent optimization or updates.

[0028] Further, in step S500, the method for collecting pest images and marking pest areas through pest monitoring is as follows: when the UAV performs a forest monitoring task, the pest monitoring degree of the real-time images is calculated. If an area is not marked as a pest area, and the occurrence monitoring degree of the pest monitoring degree of the corresponding image of the area exceeds 10%, then the area is marked as a pest area, and the obtained image is stored and recorded as a pest image; otherwise, the image is not stored.

[0029] The computer action of marking the area as a pest-infested area includes attaching the GPS location to image metadata and subsequently storing it on a server, which is then updated to a GIS layer file in the pest-infested area map.

[0030] This step, through dynamic determination of the occurrence of monitoring, controls the saving of images and the marking of areas, enabling accurate acquisition of pest images and spatial modeling of pest areas. It is a key step in the leap from UAV time perception capability to spatial intelligence capability, avoiding the storage of a large number of invalid or normal area images, saving equipment storage resources and reducing the pressure of data backhaul.

[0031] Furthermore, in step S501, the method for continuously acquiring pest monitoring data in the pest-affected area to calculate the number of forest pest occurrence trajectories is as follows:

[0032] During routine drone patrols, pest monitoring scores are obtained from the marked pest areas each day, including occurrence monitoring scores and control monitoring scores. The obtained pest monitoring scores are usually between 0 and 1. The occurrence monitoring score is the probability of determining whether pests exist, and the control monitoring score is the probability of the degree of improvement in pest control after human intervention. Both probabilities are determined based on a time-aware model.

[0033] Let a time period be defined as the monitoring period Qmet, where Qmet∈[10,30] natural days. The pest monitoring degree is obtained by the drone during the timed patrol of each natural day. The pest monitoring degree is stored as a binary array in the computer logic. The occurrence monitoring degree and the control monitoring degree are obtained within the monitoring period. The distance between the maximum and minimum values ​​and the median of the occurrence monitoring degree is calculated, and the maximum value is used as the oscillation boundary number.

[0034] The distance between the maximum and minimum values ​​and the median refers to the difference between the maximum and the median, and the difference between the median and the minimum value, respectively. The calculation of the oscillation threshold is based on the assumption that the data is sufficiently stable. By comparing the bipolar differences, the absolute difference between the median, which represents the intermediate trend, and the difference between the maximum and the minimum values ​​is calculated. The upward fluctuation and downward deviation are dynamically monitored. Taking the larger of the two values ​​as the threshold is to balance the trend and fluctuation and avoid the influence of one side. It also captures the nonlinear mutation of pest population outbreaks.

[0035] During the monitoring period, the absolute value of the difference between the prevention and control monitoring degree of any two consecutive natural days is recorded as the prevention and control step number, and the average of all prevention and control step numbers is the step average.

[0036] The step average is calculated by averaging the absolute values ​​of the changes in the prevention and control monitoring over two consecutive natural days, which measures the average degree of improvement in the prevention and control effect and provides a reference for the effectiveness of subsequent prevention and control measures.

[0037] If, when comparing a natural day with the first natural day in the opposite time direction, the increase in monitoring degree is greater than the oscillation threshold or the decrease in prevention and control monitoring degree is greater than the step average, then that natural day is recorded as a turning point marker node.

[0038] The increase in monitoring rate here refers to the difference between the monitoring rate of the current calendar day and the monitoring rate of the previous calendar day;

[0039] The principle for determining the turning marker node is that when the increase in monitoring degree is too large, it means that the risk of pest infestation is increasing, and when the decrease in control monitoring degree is too large, it means that the control measures are not effective. Based on the influence of these two variables, we can give early warning of key time nodes and further enhance the influence of these key time nodes on the assessment of the number of forest pest occurrence trajectories, so as to accurately measure the number of forest pest occurrence trajectories on the current natural day.

[0040] Set a time window pw, pw∈[5,7] natural days; the default value is 5 natural days. Calculate the difference between the monitoring degree on any natural day and pw natural days in the reverse time direction, and record it as the local change degree Lvdeg.

[0041] The short-term window was selected to analyze the recent trend of dynamic cumulative risk and eliminate long-term noise interference. The lower limit of pw is 5 days to cover the cycle of significant changes in pest activity, and the upper limit of 7 days can avoid the trend from being too smooth due to an excessively long window, thus focusing on short-term dynamics. The impact of pest situation and control measures on continuous time changes was quantified by local changes, and the short-term trend of pest changes was assessed.

[0042] By setting the penalty weighting value of the turning marker node to 1, and the rest to 0;

[0043] For the current calendar day, calculate the number of forest damage occurrence trajectories, Fsera, based on the degree of local change and the penalty weighting function:

[0044] ;

[0045] Where t1 is the cumulative variable, Lvdeg t1 Opcmr is the difference in monitoring degree between the current natural day and the t1-th natural day in the reverse time direction. t1 Let I(t1) represent the pest control monitoring rate on the t1th day in the reverse time direction from the current day. exp() is an exponential function with the natural constant e as the base; Opcmr represents the pest control monitoring rate on the current day; and I(t1) is the penalty weighting value. The weights in the penalty term are set to grow at a rate lower than linear growth, following the ecological rhythm of natural evolution where the larger the time window, the slower the daily impact weight. This avoids distortion or misjudgment of the total effect under long-term observation. If the pest control cycle is short, the penalty weighting value is set to I(t1) = 1. If the control cycle is long, requiring more than a month, the penalty weighting value is calculated using the penalty weighting function. For the t1th day, let the penalty weighting function be... Here, t1 is the sequence number of each natural day after the start of pest monitoring data collection;

[0046] The calculation principle of the forest pest occurrence trajectory number is based on the local fluctuation characteristics and intervention response trend of pest occurrence monitoring degree and control monitoring degree in continuous natural days within a short time window, simulating the interaction mechanism between "pest population dynamics - control intervention - residual rebound" in the process of forest pest development.

[0047] In natural forest ecosystems, the outbreak and decline of pests are often subject to highly time-sensitive ecological rhythms, including temperature, humidity, life cycle segments, and the delayed effects of control measures. Pest activity often exhibits nonlinear characteristics such as short-term abrupt changes and delayed responses. Therefore, by setting a near-term time window to cover the micro-periodic fluctuations of pest activity and the feedback cycle after intervention, a focused extraction of short-term pest risk dynamics can be achieved. Simultaneously, a cumulative interaction term for local change rate and daily control monitoring rate is constructed, and a cumulative mechanism with penalty weights is introduced to biomimeticly simulate the ecological process by which the accumulation of pest fluctuations drives the malignant expansion of forest pests.

[0048] It is important to understand that in natural systems, if pest outbreaks fail to be controlled within a short period, they are highly prone to rebound. For example, pests such as pine caterpillars and gypsy moths tend to regroup within a few days after being dispersed. Therefore, this method uses reversal marker nodes to identify these critical rebound points and increases the contribution of these key time points to the number of occurrence trajectories through a penalty weighting function, thus more closely resembling the stress-induced re-outbreak mechanism in natural evolution.

[0049] Since the determination of the aforementioned turning marker nodes is based on the occurrence monitoring degree and prevention monitoring degree of each time period perceived and inverted, it can accurately capture the natural days that need to be focused on when the pest intensifies or the improvement effect of prevention and control measures is poor. Therefore, it can target the short-term pest change trend and the improvement effect of prevention and control measures with penalty weighting, amplify the role of turning marker nodes, and ultimately accurately conduct dynamic assessment of the number of forest pest occurrence trajectories. However, since the number of forest pest occurrence trajectories is also closely related to the natural days with significant prevention and control effects, only amplifying the role of turning marker nodes with higher risks will reduce the accuracy of calculating the number of forest pest occurrence trajectories. At the same time, relying solely on the static comparison of thresholds to highlight key time nodes cannot dynamically reflect the residual impact of historical prevention and control effects on current risks. In order to improve the accuracy of assessing the number of forest pest occurrence trajectories, this invention proposes a better solution.

[0050] Furthermore, in step S501, the method for continuously acquiring pest monitoring data in the pest-affected area to calculate the number of forest pest occurrence trajectories is as follows:

[0051] Let a time period be defined as the monitoring period Qmet, where Qmet∈[15,30] natural days. During the timed patrol of each natural day by a drone, the pest monitoring degree is obtained, namely the occurrence monitoring degree Dgrcu and the control monitoring degree Opcmr. On any natural day within the monitoring period, the ratio of the control monitoring degree to the occurrence monitoring degree is recorded as the pest control efficiency index Ipcoe. The relative efficiency of control is quantified by the ratio of the degree of improvement in control effect to the probability of pest occurrence, which is used to evaluate the effectiveness of control measures. The higher the control efficiency index, the stronger the control effect, and vice versa, the weaker the control and the more likely the risk accumulates.

[0052] The product of the remaining control monitoring degree and the occurrence monitoring degree is the relative occurrence product Rmari, where the remaining control monitoring degree is the difference between 1 and the control monitoring degree. If there is a control monitoring degree greater than 1, the data within the monitoring period needs to be normalized. Rmari = Dgrcu × (1 - Opcmr). By calculating the nonlinear interaction between pest occurrence and control monitoring effect, the amount of pest risk that still exists after control is quantified, providing a basis for the subsequent calculation of dynamic cumulative occurrence risk.

[0053] Set the time window pw, which ranges from pw ∈ [5,7] natural days, with a default value of 5. For any given natural day, calculate the dynamic cumulative occurrence risk based on the pest control efficiency index and relative occurrence volume; Dymlr,

[0054]

[0055] Where t2 is the cumulative variable, Ipcoe t2 Rmari t2These are the pest control efficiency index and relative occurrence volume on the second natural day in the reverse time direction from the current natural day, respectively, and exp() is an exponential function with the natural constant e as the base.

[0056] The calculation principle of dynamic cumulative risk is as follows: by taking into account the influence of historical prevention and control efficiency index, the index decay weighting is used. The higher the prevention and control efficiency index of a natural day, the faster the subsequent impact decays, and therefore the smaller its contribution to the current cumulative risk. As the time span increases, the cumulative impact of risk on the current natural day decreases, so the corresponding weight decays faster, which can dynamically reflect the residual impact of historical prevention and control effects on the current risk.

[0057] For any given time window, plot a line graph of the dynamic cumulative risk. Record the day when the slope changes from negative to positive as the defect marker, and the day when it changes from positive to negative as the positive marker. Record the difference between the slope of the defect marker and the previous day as the positive marker quantity Ramdu, and the difference between the slope of the positive marker and the previous day as the defect marker quantity Ratco. Calculate the difference in dynamic cumulative risk between adjacent days using the first-order difference of the dynamic cumulative risk. This provides a time-sensitive signal for identifying turning points in risk trends. A change in the line slope from negative to positive indicates accelerated risk accumulation. The difference between positive and negative slopes quantifies the intensity of risk acceleration. Conversely, a change in slope from positive to negative indicates a slowdown in risk trends or the effectiveness of mitigation measures. Therefore, the difference between negative and positive slopes quantifies the intensity of risk mitigation and assesses the actual effectiveness of mitigation measures.

[0058] For any natural day k2, let the slope sign weighting function be denoted as... :

[0059] ;

[0060] Ramdu k2 Ratco k2 These are the positive and defect marking quantities corresponding to natural day k2, respectively. The purpose of calculating the slope sign weight function is to highlight the impact of risk inflection points by dynamically adjusting the contribution of key time nodes, using defect marking points to indicate the accelerated accumulation of risk, and amplifying its influence range by increasing the weight. Conversely, when prevention and control are improved, it indicates that the risk trend is slowing down and its influence is reduced by decreasing the weight. Here, k2 is used as the natural day sequence value.

[0061] For the current calendar day, calculate the forest damage occurrence trajectory number Fsera based on the dynamic cumulative occurrence risk and slope sign weighting function: ; where t3 is the cumulative variable, Dymlr t3 , These represent the dynamic cumulative occurrence risk and slope sign weight function value of the t3rd natural day in the counterclockwise direction from the current natural day, respectively, while Dymlr0 represents the dynamic cumulative occurrence risk and slope sign weight function value of the current natural day.

[0062] The calculation principle of the forest pest occurrence trajectory number is to simulate the natural evolution trend of the typical nonlinear process of local outbreak of pests in the forest ecosystem—control intervention—lag response—re-outbreak or mitigation by comprehensively considering the dynamic cumulative occurrence risk within the near time window, the trend slope weight function, and the time series weight decay mechanism. Its physical semantics are highly consistent with the ecological inertia in natural phenomena. That is, after pest control, due to factors such as residual insect eggs, delayed recovery of host plants, and warming climate, pest risk often exhibits a slow decline or rebound effect over time. This model effectively captures this biological lag response characteristic by assigning faster decay weights to historical time points with high control efficiency.

[0063] The model also introduces a marking mechanism based on the change in the slope direction of dynamic cumulative risk, and applies a slope sign weight function to key inflection points to characterize the sensitive time nodes of pest risk trend changes. This mechanism is essentially an abstract mathematical reconstruction of the inflection point-type outbreak or decline threshold turning point in the natural system, which can more realistically reflect the fluctuation amplification effect of risk in space and time, and form a comprehensive trend judgment of the current pest development stage.

[0064] Beneficial Effects: Since the number of forest pest occurrence trajectories is a continuous judgment result of pest monitoring degree based on the time dimension, it can effectively quantify the problem of difficulty in objectively judging the monitoring termination node due to the ambiguity of pest residual risk changes in the forest pest monitoring stage after prevention and control intervention. It reduces the inefficient execution and error risk of traditional pest control monitoring that relies heavily on manual operation. It effectively explains the degree of occurrence at the end of the control system from the rationality of the trajectory, thus forming a closed loop of the UAV monitoring process and improving the autonomy and automation capabilities of UAV forest pest monitoring. By introducing the pest control efficiency index and relative occurrence volume, and integrating time-series decay, slope change markers and weight reconstruction mechanisms, it realizes the modeling of insect population density fluctuations and control lag effects at the microscale. It uses time-series perception to strengthen the hierarchical memory of historical intervention response and inflection point sensitivity. At the macro level, it forms a structural trajectory quantification of forest pest evolution trend, providing decision-making self-evolution capability for ecological risk governance, endowing the model with time-dominated evolutionary understanding, so that pest trend judgment is no longer a static estimate, but has dynamic prediction and response discrimination capabilities, and constructs a forest pest intelligent monitoring paradigm oriented towards the prevention and control closed loop.

[0065] Furthermore, in step S502, the method for judging the forest pest monitoring result based on the number of forest pest occurrence trajectories is as follows: when the number of forest pest occurrence trajectories reaches a minimum value and the number of forest pest occurrence trajectories is higher than the minimum value for at least 15 natural days thereafter, the forest pest monitoring result for the area is determined to be forest pest elimination, and the pest area marker for the area is canceled.

[0066] Preferably, all undefined variables in this invention, if not explicitly defined, can be manually set thresholds.

[0067] This invention also provides a time-aware attention-based drone forest damage monitoring system. The time-aware attention-based drone forest damage monitoring system includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the time-aware attention-based drone forest damage monitoring method. The time-aware attention-based drone forest damage monitoring system can run on computing devices such as desktop computers, laptops, handheld computers, and cloud data centers. The runnable system may include, but is not limited to, processors, memory, and server clusters. The processor executes the computer program within the following system units:

[0068] The forest damage image sequence library construction unit is used to construct a forest damage image sequence library from images collected by drones.

[0069] The time-aware modeling unit is used to construct a time-aware attention-based pest perception model using a forest damage image sequence library to obtain pest-related vectors.

[0070] Perceptron building unit, used to build a K-layer perceptron from pest-related vectors;

[0071] The sensor data inversion unit is used to extract image descriptors from the K-layer sensor and obtain pest monitoring degree;

[0072] The pest area marking unit is used to collect pest images and mark pest areas through pest monitoring.

[0073] The forest pest occurrence trajectory number calculation unit is used to continuously acquire pest monitoring data in pest-affected areas to calculate the forest pest occurrence trajectory number.

[0074] The prevention and control feedback unit is used to determine the forest damage monitoring results based on the number of forest damage occurrence trajectories.

[0075] The beneficial effects of this invention are as follows: This invention provides a drone forest pest monitoring method and system based on time-aware attention. Through K-layer perception, it achieves lightweight and rapid judgment at the drone end, freeing up power consumption and computing power, saving storage resources and improving endurance. It prevents signal anxiety in forest scenes, effectively quantifies the problem of difficulty in objectively judging the monitoring termination node due to the ambiguity of pest residual risk changes in the forest pest monitoring stage after prevention and control intervention, reduces the inefficient execution and misoperation risk of traditional pest control monitoring that relies heavily on manual operation, effectively explains the degree of occurrence of the endpoint of the prevention and control system from the rationality of the trajectory, and thus forms a closed loop of drone monitoring process, improving the autonomy and automation capabilities of drone forest pest monitoring. Attached Figure Description

[0076] The above and other features of the present invention will become more apparent from the detailed description of the embodiments shown in conjunction with the accompanying drawings. In the accompanying drawings, the same reference numerals denote the same or similar elements. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without any creative effort. In the drawings:

[0077] Figure 1 The diagram shows a flowchart of a drone-based forest damage monitoring method based on time-aware attention.

[0078] Figure 2 The diagram shows the structure of a drone-based forest damage monitoring system based on time-aware attention. Detailed Implementation

[0079] The following will provide a clear and complete description of the concept, specific structure, and technical effects of the present invention in conjunction with the embodiments and accompanying drawings, so as to fully understand the purpose, solution, and effects of the present invention. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0080] like Figure 1 The diagram shows a flowchart of a drone-based forest damage monitoring method based on time-aware attention. The following section will combine... Figure 1 This invention describes a time-aware attention-based unmanned aerial vehicle (UAV) method for monitoring forest damage, comprising the following steps:

[0081] S100, using drones to collect images and construct a forest damage image sequence library;

[0082] S200, using a forest damage image sequence library to construct a time-aware attention-based pest perception model to obtain pest-related vectors;

[0083] S300, constructs a K-layer perceptron based on pest-related vectors;

[0084] S400 extracts image descriptors from the K-layer perceptron and obtains pest monitoring accuracy;

[0085] S500 collects pest images and marks pest-infested areas through pest monitoring;

[0086] S501, Calculate the number of forest pest occurrence trajectories by continuously acquiring pest monitoring data in pest-affected areas;

[0087] S502, determine the forest damage monitoring results based on the number of forest damage occurrence trajectories.

[0088] Furthermore, in step S100, the method for constructing a forest pest image sequence library by collecting images with a drone is as follows: the flight route and monitoring cycle are set according to the known locations of pests in the forest area, and images are collected by the drone at regular intervals every day to form image sequence instances; the sequence formed by arranging the images of various pests at the same location in chronological order constitutes an image sequence instance, and the forest pest image sequence library is composed of several pest image sequence instances.

[0089] The forest pest image sequence library corresponds to only one type of forest pest. The forest pest must have obvious visual symptoms, a clear development stage, and strong regional concentration. The types of forest pests can include pine sawyer beetle, gypsy moth, poplar inchworm, or pine caterpillar, etc.

[0090] The image sequence instance includes a start frame, an outbreak frame, and a prevention end frame. The start frame represents the time point in the image sequence where pest symptoms or suspected symptoms first appear. The outbreak frame is the time point in the image sequence where pest symptoms spread rapidly and reach a significant level. The prevention end frame is the time point in the image sequence where pest symptoms have been basically alleviated after human intervention, including spraying, using insect nets, or forest land management.

[0091] In the drone's flight path planning, the altitude is set to 30-60m, with a default value of 30m; the ground sampling interval is 3-5cm / pixel, with a default value of 5cm / pixel; the camera's field of view (FOV) is 84° horizontally and 63° vertically; the forward overlap rate is greater than or equal to 70%, with a default value of 70%; the lateral overlap rate is greater than or equal to 60%, with a default value of 60%; and the path mode is a zig-zag serpentine flight path.

[0092] Further, in step S200, the method for constructing a time-aware attention-based pest perception model using a forest damage image sequence library to obtain pest-related vectors is as follows: After preprocessing the forest damage image sequence library, a deep learning model containing a time-aware attention mechanism is used to train the image sequences to obtain the pest perception model. The time-aware attention mechanism employs any one of the TimeSformer operator, ConvLSTM operator, or Temporal Attention Network operator. The output of the intermediate embedding layer of the pest perception model is extracted as a pest-related vector, the mathematical expression of which is: , where d is the feature dimension, with a value of 128 or 256.

[0093] Preprocessing is required for any image sequence instance, including image alignment, uniform size adjustment, and color normalization.

[0094] The time-aware deep learning model uses the TimeSformer operator, ConvLSTM operator, or Temporal Attention Network operator. The default operator is TimeSformer, which is a common operator specifically designed for processing video sequences or multi-temporal image sequences. Since the development of pests is usually not linear and stable but has abrupt change nodes, the TimeSformer operator has a long-distance attention mechanism in the time dimension, which is beneficial for capturing non-continuous but causally related pest dynamics across time.

[0095] The output pest perception model is a trained pest perception model, whose prediction features are occurrence confidence and control confidence. Occurrence confidence refers to the percentile value of an image at a given time point between the start frame and the outbreak frame, and control confidence refers to the percentile value of an image at a given time point between the outbreak frame and the prevention end frame. The control confidence of each image between the start frame and the outbreak frame is 0, and the occurrence confidence of each image between the outbreak frame and the prevention end frame is 1.

[0096] During the training of the pest perception model, the loss function includes cross-entropy loss, temporal consistency regularization term and attention sparsity penalty term; the optimizer adopts Adam operator, the learning rate is set to 1e-4, the number of training rounds is 100, and the data in each round is divided into 70% of the forest pest image sequence library for training, 15% for validation, and 15% for test.

[0097] Further, in step S300, the method for constructing a K-layer perceptron based on the pest-related vector is as follows: the pest-related vector is used as input to construct a feedforward neural network model containing at least K fully connected layers as a K-layer perceptron, wherein each hidden layer adopts the ReLU activation function, and the output layer outputs including occurrence confidence and control confidence, which are used to represent the temporal position state of the input image in the pest development process.

[0098] In the structural design of the K-layer perceptron, the input layer dimension is consistent with the dimension of the pest-related vector, the number of hidden layers is K-2, K≥3, with a default value of 5, each layer adopts a fully connected structure and uses the ReLU activation function; the output layer dimension is 2, corresponding to the occurrence confidence and prevention confidence; the loss function set by the K-layer perceptron adopts the mean squared error loss, which supervises the occurrence confidence and prevention confidence respectively, with a learning rate of 1e-4, and is optimized by the Adam operator, and the generalization ability is enhanced by the Dropout and BatchNorm operators.

[0099] Further, in step S400, the method for extracting image descriptors from the K-layer perceptron and obtaining pest monitoring degree is as follows: the UAV inputs the collected images into the K-layer perceptron to obtain image descriptors and obtains the occurrence confidence and control confidence of real-time monitoring through nonlinear mapping processing, which are defined as occurrence monitoring degree and control monitoring degree, respectively, and both are classified as pest monitoring degree.

[0100] The resulting image descriptors have the same format as the pest-related vectors. As intermediate vectors within the perceptron, their mathematical expression is: , where d is the feature dimension, with a value of 128 or 256; non-linear processing uses the ReLU activation function by default.

[0101] Further, in step S500, the method for collecting pest images and marking pest areas through pest monitoring is as follows: when the UAV performs a forest monitoring task, the pest monitoring degree of the real-time images is calculated. If an area is not marked as a pest area, and the occurrence monitoring degree of the pest monitoring degree of the corresponding image of the area exceeds 10%, then the area is marked as a pest area, and the obtained image is stored and recorded as a pest image; otherwise, the image is not stored.

[0102] The computer action of marking the area as a pest-infested area includes attaching the GPS location to image metadata and subsequently storing it on a server, which is then updated to a GIS layer file in the pest-infested area map.

[0103] Furthermore, in step S501, the method for continuously acquiring pest monitoring data in the pest-affected area to calculate the number of forest pest occurrence trajectories is as follows:

[0104] During routine drone patrols, pest monitoring scores are obtained from the marked pest areas each day, including occurrence monitoring scores and control monitoring scores. The obtained pest monitoring scores are usually between 0 and 1. The occurrence monitoring score is the probability of determining whether pests exist, and the control monitoring score is the probability of the degree of improvement in pest control after human intervention. Both probabilities are determined based on a time-aware model.

[0105] Let a time period be defined as the monitoring period Qmet, where Qmet∈[10,30] natural days. The pest monitoring degree is obtained by the drone during the timed patrol of each natural day. The pest monitoring degree is stored as a binary array in the computer logic. The occurrence monitoring degree and the control monitoring degree are obtained within the monitoring period. The distance between the maximum and minimum values ​​and the median of the occurrence monitoring degree is calculated, and the maximum value is used as the oscillation boundary number.

[0106] The distances between the maximum and minimum values ​​and the median refer to the differences between the maximum and the median, and the differences between the median and the minimum value, respectively.

[0107] During the monitoring period, the absolute value of the difference between the prevention and control monitoring degree of any two consecutive natural days is recorded as the prevention and control step number, and the average of all prevention and control step numbers is the step average.

[0108] If, when comparing a natural day with the first natural day in the opposite time direction, the increase in monitoring degree is greater than the oscillation threshold or the decrease in prevention and control monitoring degree is greater than the step average, then that natural day is recorded as a turning point marker node.

[0109] The increase in monitoring rate here refers to the difference between the monitoring rate of the current calendar day and the monitoring rate of the previous calendar day;

[0110] Set a time window pw, pw∈[5,7] natural days; the default value is 5 natural days. Calculate the difference between the monitoring degree on any natural day and pw natural days in the reverse time direction, and record it as the local change degree Lvdeg.

[0111] By setting the penalty weighting value of the turning marker node to 1, and the rest to 0;

[0112] For the current calendar day, calculate the number of forest damage occurrence trajectories, Fsera, based on the degree of local change and the penalty weighting function:

[0113] ;

[0114] Where t1 is the cumulative variable, Lvdeg t1 Opcmr is the difference in monitoring degree between the current natural day and the t1-th natural day in the reverse time direction. t1Let I(t1) represent the pest control monitoring rate on the t1th day in the reverse time direction from the current day. exp() is an exponential function with the natural constant e as the base; Opcmr represents the pest control monitoring rate on the current day; and I(t1) is the penalty weighting value. The weights in the penalty term are set to grow at a rate lower than linear growth, following the ecological rhythm of natural evolution where the larger the time window, the slower the daily impact weight. This avoids distortion or misjudgment of the total effect under long-term observation. If the pest control cycle is short, the penalty weighting value is set to I(t1) = 1. If the control cycle is long, requiring more than a month, the penalty weighting value is calculated using the penalty weighting function. For the t1th day, let the penalty weighting function be... Here, t1 is the sequence number of each natural day after the start of pest monitoring data collection;

[0115] The calculation principle of the forest pest occurrence trajectory number is based on the local fluctuation characteristics and intervention response trend of pest occurrence monitoring degree and control monitoring degree in continuous natural days within a short time window, simulating the interaction mechanism between "pest population dynamics - control intervention - residual rebound" in the process of forest pest development.

[0116] In natural forest ecosystems, the outbreak and decline of pests are often subject to highly time-sensitive ecological rhythms, including temperature, humidity, life cycle segments, and the delayed effects of control measures. Pest activity often exhibits nonlinear characteristics such as short-term abrupt changes and delayed responses. Therefore, by setting a near-term time window to cover the micro-periodic fluctuations of pest activity and the feedback cycle after intervention, a focused extraction of short-term pest risk dynamics can be achieved. Simultaneously, a cumulative interaction term for local change rate and daily control monitoring rate is constructed, and a cumulative mechanism with penalty weights is introduced to biomimeticly simulate the ecological process by which the accumulation of pest fluctuations drives the malignant expansion of forest pests.

[0117] It is important to understand that in natural systems, if pest outbreaks fail to be controlled within a short period, they are highly prone to rebound. For example, pests such as pine caterpillars and gypsy moths tend to regroup within a few days after being dispersed. Therefore, this method uses reversal marker nodes to identify these critical rebound points and increases the contribution of these key time points to the number of occurrence trajectories through a penalty weighting function, thus more closely resembling the stress-induced re-outbreak mechanism in natural evolution.

[0118] Furthermore, in step S501, the method for continuously acquiring pest monitoring data in the pest-affected area to calculate the number of forest pest occurrence trajectories is as follows:

[0119] Let a time period be designated as the monitoring period Qmet, where Qmet ∈ [15, 30] natural days. During the timed patrol of each natural day by a drone, the pest monitoring degree is obtained, namely the occurrence monitoring degree Dgrcu and the control monitoring degree Opcmr. On any natural day within the monitoring period, the ratio of the control monitoring degree to the occurrence monitoring degree is denoted as the pest control efficiency index Ipcoe.

[0120] The product of the residual control monitoring degree and the occurrence monitoring degree is the relative occurrence product Rmari, where the residual control monitoring degree is the difference between 1 and the control monitoring degree. If there is a control monitoring degree greater than 1, the data within the monitoring period needs to be normalized; Rmari = Dgrcu × (1 - Opcmr);

[0121] Set the time window pw, which ranges from pw ∈ [5,7] natural days, with a default value of 5. For any given natural day, calculate the dynamic cumulative occurrence risk based on the pest control efficiency index and relative occurrence volume; Dymlr,

[0122]

[0123] Where t2 is the cumulative variable, Ipcoe t2 Rmari t2 These are the pest control efficiency index and relative occurrence volume on the second natural day in the reverse time direction from the current natural day, respectively, and exp() is an exponential function with the natural constant e as the base.

[0124] Plot a line graph of the dynamic cumulative risk for any time window. Record the day corresponding to the point where the slope changes from negative to positive as the defect marker point, and the day corresponding to the point where the slope changes from positive to negative as the positive marker point. Record the difference between the slope of the defect marker and the previous day as the positive marker quantity Ramdu, and record the difference between the slope of the positive marker point and the previous day as the defect marker quantity Ratco.

[0125] For any natural day k2, let the slope sign weighting function be denoted as... :

[0126] ;

[0127] Ramdu k2 Ratco k2 These are the positive and defective marker values ​​corresponding to natural day k2, respectively.

[0128] For the current calendar day, calculate the forest damage occurrence trajectory number Fsera based on the dynamic cumulative occurrence risk and slope sign weighting function: ; where t3 is the cumulative variable, Dymlr t3 , These represent the dynamic cumulative occurrence risk and slope sign weight function value of the t3rd natural day in the counterclockwise direction from the current natural day, respectively, while Dymlr0 represents the dynamic cumulative occurrence risk and slope sign weight function value of the current natural day.

[0129] Furthermore, in step S502, the method for judging the forest pest monitoring result based on the number of forest pest occurrence trajectories is as follows: when the number of forest pest occurrence trajectories reaches a minimum value and the number of forest pest occurrence trajectories is higher than the minimum value for at least 15 natural days thereafter, the forest pest monitoring result for the area is determined to be forest pest elimination, and the pest area marker for the area is canceled.

[0130] An embodiment of the present invention provides a time-aware attention-based unmanned aerial vehicle (UAV) forest damage monitoring system, such as... Figure 2 The diagram shows a structure of a time-aware attention-based drone forest damage monitoring system according to the present invention. This embodiment of the time-aware attention-based drone forest damage monitoring system includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the above embodiment of the time-aware attention-based drone forest damage monitoring method.

[0131] The system includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program in units of the following system:

[0132] The forest damage image sequence library construction unit is used to construct a forest damage image sequence library from images collected by drones.

[0133] The time-aware modeling unit is used to construct a time-aware attention-based pest perception model using a forest damage image sequence library to obtain pest-related vectors.

[0134] Perceptron building unit, used to build a K-layer perceptron from pest-related vectors;

[0135] The sensor data inversion unit is used to extract image descriptors from the K-layer sensor and obtain pest monitoring degree;

[0136] The pest area marking unit is used to collect pest images and mark pest areas through pest monitoring.

[0137] The forest pest occurrence trajectory number calculation unit is used to continuously acquire pest monitoring data in pest-affected areas to calculate the forest pest occurrence trajectory number.

[0138] The prevention and control feedback unit is used to determine the monitoring results of forest damage based on the number of forest damage occurrence trajectories.

[0139] The aforementioned time-aware attention-based drone-based forest damage monitoring system can run on computing devices such as desktop computers, laptops, handheld computers, and cloud servers. The system that can run on this time-aware attention-based drone-based forest damage monitoring system may include, but is not limited to, processors and memory. Those skilled in the art will understand that the example described is merely an illustration of a time-aware attention-based drone-based forest damage monitoring system and does not constitute a limitation on such a system. It may include more or fewer components, or a combination of certain components, or different components. For example, the aforementioned time-aware attention-based drone-based forest damage monitoring system may also include input / output devices, network access devices, buses, etc.

[0140] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. This processor is the control center of the time-aware attention-based UAV forest damage monitoring system, connecting various parts of the system via various interfaces and lines.

[0141] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the time-aware attention-based UAV forest damage monitoring system by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created based on the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0142] Although the invention has been described in considerable detail and particularly with regard to several of the described embodiments, it is not intended to limit itself to any of these details or embodiments or any particular embodiment, thereby effectively covering the intended scope of the invention. Furthermore, the invention has been described above with respect to embodiments foreseeable by the inventors in order to provide a useful description, and non-substantial modifications to the invention that have not yet been foreseen may still represent equivalent modifications.

Claims

1. A method for monitoring forest damage using unmanned aerial vehicles (UAVs) based on time-aware attention, characterized in that, The method includes the following steps: S100, using drones to collect images and construct a forest damage image sequence library; S200, using a forest damage image sequence library to construct a time-aware attention-based pest perception model to obtain pest-related vectors; S300, constructs a K-layer perceptron based on pest-related vectors; S400 extracts image descriptors from the K-layer perceptron and obtains pest monitoring accuracy; S500 collects pest images and marks pest-infested areas through pest monitoring; S501, Calculate the number of forest pest occurrence trajectories by continuously acquiring pest monitoring data in pest-affected areas; S502, determine the forest damage monitoring results based on the number of forest damage occurrence trajectories; In step S400, the method for extracting image descriptors from the K-layer perceptron and obtaining pest monitoring degree is as follows: the UAV inputs the acquired images into the K-layer perceptron to obtain image descriptors, and obtains the occurrence confidence degree and prevention confidence degree of real-time monitoring through nonlinear mapping processing, which are defined as occurrence monitoring degree and prevention monitoring degree, respectively, and both are classified as pest monitoring degree; where occurrence confidence degree refers to the percentile value of an image at a corresponding time point between the start frame and the outbreak frame, and prevention confidence degree refers to the percentile value of an image at a corresponding time point between the outbreak frame and the prevention end frame; The method for calculating the number of forest pest occurrence trajectories by continuously acquiring pest monitoring information in pest-affected areas in S501 is as follows: the oscillation boundary number is obtained by horizontal comparison of occurrence monitoring information in pest monitoring information; the difference between control monitoring information of two consecutive natural days is obtained as the control step number, the average level of which is the step average quantity; the oscillation boundary number and the step average quantity are used to mark the turning marker node, thereby setting the penalty weighting value, and calculating the degree of change based on the occurrence monitoring information.

2. The method for monitoring forest damage by unmanned aerial vehicles based on time-aware attention according to claim 1, characterized in that, In step S100, the method for constructing a forest pest image sequence library by collecting images with a drone is as follows: the flight route and monitoring cycle are set according to the known locations of pests in the forest area, and images are collected by the drone at regular intervals every day to form image sequence instances; the sequence formed by arranging the images of various pests at the same location in chronological order constitutes an image sequence instance, and the forest pest image sequence library is composed of several pest image sequence instances.

3. The method for monitoring forest damage by unmanned aerial vehicles based on time-aware attention according to claim 1, characterized in that, In step S200, the method for constructing a time-aware attention-based pest perception model using a forest damage image sequence library to obtain pest-related vectors is as follows: after preprocessing the forest damage image sequence library, a deep learning model containing a time-aware attention mechanism is used to train the image sequence to obtain the pest perception model. The time-aware attention mechanism adopts any one of the TimeSformer operator, ConvLSTM operator, or Temporal Attention Network operator. The output of the intermediate embedding layer of the pest perception model is extracted as the pest-related vector.

4. The method for monitoring forest damage by unmanned aerial vehicles based on time-aware attention according to claim 1, characterized in that, In step S300, the method for constructing a K-layer perceptron based on pest-related vectors is as follows: the pest-related vectors are used as input to construct a feedforward neural network model containing at least K fully connected layers as a K-layer perceptron, wherein each hidden layer adopts the ReLU activation function, and the output layer outputs occurrence confidence and control confidence, which are used to represent the temporal position state of the input image in the pest development process.

5. The method for monitoring forest damage by unmanned aerial vehicles based on time-aware attention according to claim 1, characterized in that, In step S500, the method for collecting pest images and marking pest areas through pest monitoring is as follows: when the UAV performs a forest monitoring task, the pest monitoring degree of the real-time images is calculated. If an area is not marked as a pest area, and the occurrence monitoring degree of the pest monitoring degree of the corresponding image of the area exceeds 10%, then the area is marked as a pest area, and the obtained image is stored and recorded as a pest image; otherwise, the image is not stored.

6. The method for monitoring forest damage by unmanned aerial vehicles based on time-aware attention according to claim 1, characterized in that, In step S501, the method for continuously acquiring pest monitoring information in the pest-affected area to calculate the number of forest pest occurrence trajectories is as follows: acquire the occurrence monitoring information and control monitoring information during the monitoring period, calculate the distance between the maximum and minimum values ​​and the median of the occurrence monitoring information, and use the maximum value as the oscillation boundary number; during the monitoring period, record the absolute value of the difference between the control monitoring information of any two consecutive natural days as the control step number, and then the average of all control step numbers is the step average. If, when comparing a natural day with the first natural day in the opposite time direction, the increase in monitoring degree is greater than the oscillation threshold or the decrease in prevention and control monitoring degree is greater than the step average, then that natural day is recorded as a turning point marker node; set a time window pw, and calculate the difference in monitoring degree between any natural day and pw natural days in the opposite time direction, which is recorded as the local change degree Lvdeg; By setting the penalty weighting value of the turning marker node to 1 and the rest to 0, the number of forest damage occurrence trajectories is calculated for the current natural day based on the degree of local change and the penalty weighting value.

7. The method for monitoring forest damage by unmanned aerial vehicles based on time-aware attention according to claim 1, characterized in that, In step S501, the method for continuously acquiring pest monitoring information in the pest-affected area to calculate the number of forest pest occurrence trajectories is as follows: on any natural day within the monitoring period, the ratio of the control monitoring information to the occurrence monitoring information is recorded as the pest control efficiency index. The product of the remaining control monitoring degree and the occurrence monitoring degree is calculated as the relative occurrence accumulation; a time window pw is set, and the dynamic cumulative occurrence risk is calculated for any natural day based on the pest control efficiency index and the relative occurrence accumulation; For any given time window, plot a line graph of the dynamic cumulative occurrence risk. Record the day corresponding to the point where the slope changes from negative to positive as the defect marker point, and the day corresponding to the point where the slope changes from positive to negative as the positive marker point. Record the difference between the defect marker and the slope of the previous day as the positive marker quantity, and record the difference between the positive marker point and the slope of the previous day as the defect marker quantity. Calculate the slope sign weight for any given day and combine it with the dynamic cumulative occurrence risk to calculate the number of forest damage occurrence trajectories.

8. The method for monitoring forest damage by unmanned aerial vehicles based on time-aware attention according to claim 1, characterized in that, In step S502, the method for judging the forest pest monitoring result based on the number of forest pest occurrence trajectories is as follows: when the number of forest pest occurrence trajectories reaches a minimum value and the number of forest pest occurrence trajectories is higher than the minimum value for at least 15 natural days thereafter, the forest pest monitoring result for the area is determined to be forest pest elimination, and the pest area marker for the area is canceled.

9. A drone-based forest damage monitoring system based on time-aware attention, characterized in that, The time-aware attention-based drone forest damage monitoring system includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the time-aware attention-based drone forest damage monitoring method according to any one of claims 1-8. The time-aware attention-based drone forest damage monitoring system runs on a desktop computer, laptop computer, handheld computer, or cloud data center computing device.

Images