Farmland pest intelligent monitoring and migratory flight early warning method and system

Abstract

The invention discloses a farmland pest intelligent monitoring and migratory flight early warning method and system, and belongs to the technical field of agricultural pest monitoring, and the method comprises a multi-source data fusion step, a pest identification step, a dynamic analysis step, a population prediction step and a migratory flight early warning step. By integrating insect trapping images of insect situation forecasting lamps, sex attractant counting data, meteorological data, insect radar data and a pest biological characteristic knowledge base, a deep learning target detection network is adopted to realize pest recognition, peak hours and generation times are analyzed based on a down-and-long curve, a meteorological coupling time sequence prediction network is adopted to predict population trends, and the pest detection accuracy is improved. And radar and meteorological data are fused to realize migration flight early warning.

Description

Technical Field

[0001] This invention relates to the field of agricultural pest monitoring and early warning technology, and in particular to intelligent monitoring and migration early warning methods and systems for farmland pests. Background Technology

[0002] Farmland pest monitoring and early warning are crucial aspects of agricultural production, and accurate and timely pest monitoring data plays a key role in scientific prevention and control decisions. Existing pest monitoring technologies mainly rely on manual field surveys and traditional light trapping and counting methods, which suffer from problems such as high workload, poor timeliness, and strong subjectivity.

[0003] Chinese patent CN116596873A discloses an artificial intelligence-based crop pest and disease early warning system. This system includes a crop image data acquisition module, an environmental data acquisition module, a pest and disease data extraction module, a pest and disease quantity analysis module, a pest and disease species analysis module, a pest situation analysis module, and an early warning system. This solution acquires crop image data and analyzes the pest and disease images using a Mask R-CNN instance segmentation network to obtain information on the quantity and species of pests and diseases. Combined with environmental data, it obtains pest situation information and issues early warnings. However, this scheme still has the following shortcomings: First, it only uses a single image data source for pest identification and lacks the ability to integrate and process multi-source monitoring data such as images from insect monitoring lamps and counts from pheromone traps, resulting in limited comprehensiveness and accuracy of the monitoring data. Second, it lacks the ability to analyze the dynamics of pest populations, and cannot identify the peak periods and generations of pests, making it difficult to support precise decision-making on control timing. Third, it lacks the ability to predict the trends of pest populations and cannot predict the development of pest situations in the future. Finally, it completely lacks the ability to monitor and warn of migratory pests, and cannot use insect radar data to analyze the migration paths and landing areas of migratory pests, resulting in a serious lag in early warning of major migratory pests such as fall armyworm, rice planthopper, and armyworm.

[0004] In recent years, deep learning technology has made significant progress in the field of agricultural pest identification. Target detection algorithms based on convolutional neural networks can automatically identify various pests with high accuracy. Meanwhile, the development of insect radar technology has provided new means for monitoring migratory pests. Vertical observation insect radar can detect individual insects flying at high altitudes and obtain parameters such as their flight altitude, speed, and direction. However, how to effectively integrate multi-source heterogeneous monitoring data and construct an intelligent monitoring system that combines pest identification, dynamic analysis, trend prediction, and migration early warning functions remains a pressing technical challenge. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for intelligent monitoring and early warning of agricultural pests migration, which solves technical problems such as insufficient fusion of multi-source monitoring data, lack of population dynamic analysis and trend prediction capabilities, and delayed early warning of migratory pests.

[0006] The first aspect of this invention provides a method for intelligent monitoring and early warning of agricultural pests migration, including a multi-source data fusion step, a pest identification step, an occurrence dynamic analysis step, a population prediction step, and a migration early warning step. The multi-source data fusion step acquires insect-attracting image data collected by intelligent insect monitoring lamps, pest count data collected by pheromone traps, meteorological monitoring data collected by meteorological monitoring stations, radar insect monitoring data collected by insect radar, and pest biological characteristic data from a pest biological characteristic knowledge base. The above data are then spatiotemporally aligned to generate a fused dataset. The pest identification step inputs the insect-attracting image data from the fused dataset into a multi-scale feature fusion target detection network, outputting pest identification results including pest species, pest quantity, and identification confidence level. The occurrence dynamic analysis step generates a daily insect-attracting quantity sequence based on the pest identification results and pest count data, plots pest growth and decline curves, identifies peak periods, and determines the generation of pests. The population prediction step inputs pest population dynamics curves, peak periods, generations, and meteorological monitoring data into a meteorological coupled time-series prediction network to predict pest population trends within a preset time window. The migration warning step extracts upper-altitude insect flight parameters from radar insect monitoring data, combines them with upper-air meteorological data to perform migration trajectory inversion, generates migration warning information, and feeds the population prediction results back to the occurrence dynamic analysis step to dynamically adjust the peak period determination parameters.

[0007] Preferably, in the pest identification step, when the identification confidence level is lower than a preset confidence level threshold, the corresponding insect-attracting image data is marked as a sample to be reviewed, and the confidence level threshold ranges from 0.85 to 0.95.

[0008] Preferably, in the dynamic analysis step, the peak detection adopts an adaptive threshold method. When the number of insects attracted in a single day of the pest fluctuation curve exceeds a preset multiple of the average value of the previous preset number of days, it is determined to be the start of the peak period. The preset number of days is 5 to 10 days, and the preset multiple is 1.5 to 3.0 times.

[0009] Preferably, in the population prediction step, the preset time window is 7 to 14 days, and the input of the meteorological coupled time series prediction network includes the pest growth and decline curves of the preset historical days and the temperature, humidity, precipitation and wind speed data of the corresponding time period, with the preset historical days being 14 to 30 days.

[0010] Preferably, the multi-scale feature fusion target detection network includes a feature extraction backbone network, a multi-scale feature pyramid network, and a detection head network.

[0011] Preferably, the meteorological coupled time-series prediction network includes an encoder and a decoder. The encoder uses a gated recurrent unit to encode the time-series features of the pest growth curve and uses an attention mechanism to fuse meteorological monitoring data to generate a context vector.

[0012] Preferably, the flight parameters of high-altitude insects include flight altitude, flight speed, flight direction, and radar cross-section.

[0013] Preferably, the spatiotemporal alignment processing includes sorting the insect-attracting image data by the acquisition timestamp, resampling the meteorological monitoring data at a preset time interval, and performing coordinate transformation on the radar insect monitoring data, with the preset time interval being 5 to 30 minutes.

[0014] Preferably, the biological characteristics data of the pests include the developmental starting temperature, effective accumulated temperature, phototactic wavelength response characteristics, migration altitude range, and migration season.

[0015] The second aspect of the present invention provides an intelligent monitoring and migration early warning system for farmland pests, including a multi-source data fusion module, a pest identification module, an occurrence dynamic analysis module, a population prediction module, and a migration early warning module, with each module corresponding to a corresponding step in the above method.

[0016] The beneficial effects of this invention are as follows:

[0017] First, this invention integrates multi-source heterogeneous data, such as insect images from insect monitoring lamps, counting data from pheromone traps, meteorological monitoring data, insect radar data, and a knowledge base of pest biological characteristics, to achieve comprehensive fusion of monitoring data. Compared with monitoring schemes based on a single data source, the overall accuracy of pest identification is increased to over 90%.

[0018] Secondly, this invention constructs a complete analysis chain from pest identification to dynamic analysis and trend prediction, which can automatically identify the peak period and generation of pests and predict the population trend in the next 7 to 14 days, providing a scientific basis for precise prevention and control decisions.

[0019] Third, this invention innovatively integrates insect radar monitoring data and upper-air meteorological data, enabling early warning of the migration paths and landing areas of migratory pests. The warning time can be 24 to 72 hours in advance, significantly improving the emergency response capability for major migratory pests such as fall armyworm, rice planthopper, and armyworm.

[0020] Fourth, this invention establishes a feedback mechanism from population prediction results to dynamic analysis of occurrence, which can dynamically adjust the peak period determination parameters based on the prediction results, forming a closed-loop optimized intelligent monitoring system. Attached Figure Description

[0021] Figure 1This is a flowchart of the intelligent monitoring and migration early warning method for farmland pests of the present invention.

[0022] Figure 2 This is an architecture diagram of the intelligent monitoring and migration early warning system for farmland pests of this invention. Detailed Implementation

[0023] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby providing a clearer and more explicit definition of the scope of protection of the present invention.

[0024] Reference Figure 1 This invention provides an intelligent monitoring and migration early warning method for farmland pests. This method addresses the problems of time-consuming and labor-intensive manual surveys for farmland pest monitoring and the lag in early warning for migratory pests. It achieves intelligent pest identification, dynamic analysis, trend prediction, and migration early warning through multi-source data fusion and deep learning technology. In one embodiment of this invention, the method mainly includes a multi-source data fusion step, a pest identification step, an occurrence dynamic analysis step, a population prediction step, and a migration early warning step. These steps form a synergistic relationship of data transmission and feedback adjustment.

[0025] Step S1: Multi-source data fusion step.

[0026] The multi-source data fusion step is a fundamental step in the method of this invention, responsible for acquiring and integrating heterogeneous data from various monitoring devices and data sources. In the implementation of this invention, this step first acquires insect-attracting image data from intelligent insect monitoring lamps. These lamps use black light or frequency-vibration insecticidal lamps as the insect-attracting light source and are equipped with high-definition industrial cameras to capture images of pests falling into the insect collection chamber. Preferably, the industrial camera used has a resolution of no less than 1920×1080 pixels, and the acquisition frequency is set to one frame every 30 minutes, accumulating approximately 48 insect-attracting images per day. Simultaneously, this step acquires pest count data from pheromone traps. These pheromone traps are designed for specific pest species, such as pheromone traps for rice stem borers or fall armyworms, and are equipped with infrared counting sensors to automatically count the number of pests trapped.

[0027] Acquiring meteorological monitoring data is a crucial component of multi-source data fusion. In this embodiment, a meteorological monitoring station is deployed within the farmland monitoring area to collect real-time meteorological elements such as temperature, relative humidity, wind speed, wind direction, precipitation, and light intensity. The meteorological data collection frequency is set to once every 5 minutes to ensure the capture of short-term changes in meteorological conditions. Furthermore, this step also acquires radar insect monitoring data from an insect radar. The insect radar employs a vertical observation mode, operates at an X-band frequency of 9.4 GHz, and is capable of detecting individual insects flying at altitudes ranging from 150 meters to 2500 meters above the ground. The radar system outputs data including parameters such as target echo intensity, flight altitude, flight speed, flight direction, and radar cross-section.

[0028] The data on the biological characteristics of pests are derived from a pre-constructed knowledge base of pest biological characteristics. This knowledge base includes biological characteristic parameters of more than 100 major agricultural pests, including developmental threshold temperature, effective accumulated temperature, generation cycle, phototactic wavelength response characteristics, and migration characteristics. Taking the fall armyworm as an example, its developmental threshold temperature is 12.0℃, the effective accumulated temperature during the egg stage is 35.3 days, the effective accumulated temperature during the larval stage is 259.4 days, and the adults exhibit obvious migratory characteristics, with a migration altitude range of 300 meters to 1500 meters, and the main migration season is from April to October each year.

[0029] To address the aforementioned multi-source heterogeneous data, this invention designs a spatiotemporal alignment processing algorithm to achieve standardized data fusion. The core of spatiotemporal alignment processing lies in unifying data with different acquisition frequencies, timestamps, and spatial coordinates under the same spatiotemporal reference. Specifically, for insect-attracting image data, it is sorted according to the acquisition timestamp and a time index is established; for meteorological monitoring data, a linear interpolation method is used to resample at preset time intervals to align with the temporal granularity of the insect-attracting image data. In this embodiment, the preset time interval is set to 5 to 30 minutes, preferably 30 minutes, meaning the meteorological data is resampled to match the acquisition frequency of the insect-attracting images. For radar insect monitoring data, coordinate transformation is required to align with the geographical coordinates of the monitoring station. The raw radar data is represented in a polar coordinate system, including azimuth, elevation, and slant range, which needs to be converted to a rectangular coordinate system with the monitoring station as the origin.

[0030] The multi-source data spatiotemporal alignment algorithm proposed in this invention can be implemented using the following formula. For temporal resampling of meteorological data, let the original meteorological data sequence be... ,in For timestamps, These are meteorological element values, with the target time point being... Then the resampled meteorological element values The calculation formula is:

[0031] ,

[0032] in, For the target time point The corresponding meteorological element interpolation results, The most recent raw meteorological element values ​​before the target time point. The most recent raw meteorological element values ​​after the target time point. for The corresponding original timestamp, for The corresponding original timestamp satisfies .

[0033] For coordinate transformation of radar data, let the position of the insect target detected by the radar in the polar coordinate system be... ,in Slope distance It is the azimuth angle. If the elevation angle is used, then the position is converted to a Cartesian coordinate system with the monitoring station as the origin. The calculation formula is:

[0034] ,

[0035] ,

[0036] ,

[0037] in, For the eastward coordinate components, For northward coordinate components, For the vertical height component, For radar slant range detection, It is the azimuth angle, increasing clockwise from true north as 0 degrees. The angle of elevation is positive with the horizontal plane as 0 degrees and upward.

[0038] After spatiotemporal alignment, the multi-source data were integrated into a fused dataset. The fused dataset is stored using a unified data structure, with each record containing a timestamp, insect-attracting image file path, pheromone trapping count, meteorological element vector, radar target list, and associated pest biological characteristic parameters. This structured data organization provides a standardized data input interface for subsequent pest identification, dynamic analysis, and forecasting and early warning.

[0039] In the implementation of this invention, the multi-source data fusion module also needs to perform data quality control. For insect-attracting image data, the system automatically detects the image's exposure and sharpness. When the image is too dark or too bright, adaptive contrast adjustment is performed. When the image is blurry, it is marked as a low-quality image and its weight in subsequent analysis is reduced. For meteorological monitoring data, the system detects the completeness and rationality of the data. When missing data is detected, interpolation between previous and subsequent time points is used to fill in the missing data. When obvious outliers are detected, they are removed or corrected. For radar insect monitoring data, the system performs clutter filtering. By setting upper and lower thresholds for the radar cross-section, the system filters out echo signals from non-insect targets. The lower threshold is used to filter out echoes from small particles, and the upper threshold is used to filter out echoes from large flying animals such as birds. In this embodiment, the effective range of the radar cross-section is set to 0.01 square centimeters to 10 square centimeters, which can cover the body shape characteristics of most agricultural pests.

[0040] Furthermore, the multi-source data fusion module implements redundant data storage and backup mechanisms. All raw data is stored locally before spatiotemporal alignment processing, and the fused dataset is synchronously uploaded to the cloud data center. This dual storage mechanism ensures data security and traceability, facilitating subsequent historical data analysis and model training updates.

[0041] Step S2, Pest Identification Step.

[0042] The pest identification step is the core of this invention for achieving intelligent pest monitoring. This step employs a deep learning-based multi-scale feature fusion target detection network to automatically identify the species and quantity of pests from the attracted images. In the implementation of this invention, the design of the multi-scale feature fusion target detection network fully considers the characteristics of the attracted images: significant differences in pest size, with small pests such as aphids and leafhoppers occupying only a few dozen pixels in the image, while large pests such as adult noctuid moths can occupy hundreds of pixels; diverse pest postures, with the same species potentially exhibiting different postures such as outstretched wings, folded wings, or lying on their side; and occlusion and overlap among pests.

[0043] The multi-scale feature fusion object detection network consists of three main components: a feature extraction backbone network, a multi-scale feature pyramid network, and a detection head network. The feature extraction backbone network uses an improved ResNet-50 structure as its basic architecture, extracting multi-level feature representations of the image by stacking residual blocks. The backbone network outputs feature maps at three different scales: the first-scale feature map is 1 / 8 the size of the input image, with 256 channels, mainly containing shallow texture features; the second-scale feature map is 1 / 16 the size of the input image, with 512 channels, containing mid-level structural features; and the third-scale feature map is 1 / 32 the size of the input image, with 1024 channels, containing deep semantic features.

[0044] The multi-scale feature pyramid network performs bidirectional feature fusion on feature maps at the three scales mentioned above. The top-down fusion path first upsamples the third-scale feature map by a factor of 2 and then fuses it element-wise with the second-scale feature map; subsequently, the fused result is upsampled again by a factor of 2 and fused with the first-scale feature map. The bottom-up fusion path downsamples the feature maps using a convolution operation with a stride of 2, transferring low-scale features to high-scale features. Through this bidirectional feature fusion mechanism, each scale's feature map can simultaneously contain shallow detail information and deep semantic information, thereby improving the detection capability for pests of different sizes.

[0045] The multi-scale feature fusion algorithm designed in this invention can be described by the following formula. Let the original feature maps of the first scale, second scale, and third scale be respectively... , , The feature map after top-down fusion is , , The fusion process is as follows:

[0046] ,

[0047] ,

[0048] ,

[0049] in, This is a third-scale fused feature map. This is the second-scale fused feature map. This is the first-scale fused feature map. This indicates that a 1×1 convolution operation is used for channel number adjustment. This indicates a 2x nearest neighbor upsampling operation. , , These are the original feature maps output by the backbone network.

[0050] The detection head network outputs the bounding box coordinates and class probabilities of pests based on the fused multi-scale feature maps. For each scale of feature map, the detection head network includes a classification branch and a regression branch: the classification branch outputs the probability that each preset anchor box belongs to each pest category through multi-layer convolution and a sigmoid activation function; the regression branch outputs the offset of the bounding box relative to the anchor box. In this embodiment, the detection network supports the identification and classification of more than 100 major agricultural pests, including rice stem borers such as the rice stem borer, rice leaf roller, and rice planthopper, rice planthopper such as the brown planthopper, white-backed planthopper, and gray planthopper, noctuid moths such as armyworm, fall armyworm, beet armyworm, and cotton bollworm, as well as other important pests such as rice leaf roller, rice thrips, and aphids.

[0051] During network training, Focal Loss is used as the classification loss function to address the pest class imbalance problem, and CIoU Loss is used as the regression loss function to improve bounding box localization accuracy. Total loss function. The calculation formula is:

[0052] ,

[0053] in, For total training losses, This is the classification loss, also known as Focal Loss. The regression loss is called CIoU Loss. This is the classification loss weight coefficient with a value of 1.0. This is the regression loss weight coefficient with a value of 2.0.

[0054] Classification loss Using Focal Loss, the calculation formula is as follows:

[0055] ,

[0056] in, For classification loss value, This represents the model's predicted probability for the true class. The class balance factor is used to handle the imbalance between positive and negative samples and has a value of 0.25. The focusing parameter is used to reduce the weight of easily classified samples and has a value of 2.0.

[0057] The pest identification results output by the pest identification step include three elements: pest species, pest quantity, and identification confidence level. Pest species refers to the category label of each detected pest; pest quantity refers to the number of individuals of each pest detected; and identification confidence level indicates the reliability of each detection result. When the identification confidence level is lower than a preset confidence threshold, the corresponding insect-attracting image data is marked as a sample to be reviewed, so that it can be manually verified by professionals later. The confidence threshold ranges from 0.85 to 0.95. In this embodiment, the confidence threshold is preferably set to 0.90, meaning that identification results with a confidence level lower than 0.90 will be marked as requiring review. After actual testing and verification, the pest identification network of this invention achieved an average identification accuracy of 92.3% for 100 major pests on a test set containing 50,000 labeled insect-attracting images, with an identification accuracy of 94.7% for rice stem borer and 93.5% for fall armyworm.

[0058] In the implementation of this invention, constructing the training dataset for the multi-scale feature fusion target detection network is crucial to ensuring recognition accuracy. This invention collected approximately 200,000 original images of insect-attracting lamps from 100 monitoring stations across 25 provinces in China over three consecutive years. Professional insect taxonomists manually labeled the pests in the images, including the bounding box coordinates and category labels. To improve the model's adaptability to different lighting conditions and background environments, various data augmentation strategies were employed during training, including random horizontal flipping, random rotation, random brightness adjustment, random contrast adjustment, and random cropping and scaling. Furthermore, to address the issue of insufficient sample numbers for some rare pest species, oversampling and synthetic sample generation techniques were used for data balancing.

[0059] The inference process of the multi-scale feature fusion target detection network has been optimized for edge deployment scenarios. In this embodiment, model quantization technology is used to convert the floating-point model into an INT8 fixed-point model, compressing the model size from the original 98MB to 25MB, increasing the inference speed by approximately 3 times, while reducing the recognition accuracy by only 0.3 percentage points. The quantized model can be deployed on intelligent insect monitoring lamps equipped with edge computing chips to achieve local real-time recognition and reduce reliance on cloud servers. When the edge device detects suspected high-harm insects or when the recognition confidence level is low, the system automatically uploads the original image to the cloud for verification and recognition. This edge-cloud collaborative architecture balances real-time recognition and accuracy.

[0060] Step S3: Dynamic analysis steps occur.

[0061] The dynamic analysis step is a key step in this invention, transforming pest monitoring from static identification to dynamic analysis. This step generates a daily insect attraction sequence based on pest identification results and pheromone trap count data, plots pest population dynamics curves, and further identifies peak periods and generations of pest occurrence. In the implementation of this invention, this step first summarizes and statistically analyzes the daily pest identification results, then performs data correction and fusion by combining the pheromone trap count data. Insect monitoring lamps primarily attract pests positively attracted to light, while pheromone traps exhibit high species specificity for male adults of specific pest species. The fusion of these two data sources improves the accuracy and comprehensiveness of insect attraction statistics.

[0062] The process of generating the daily insect trapping sequence is as follows: For each monitoring date First, the number of each type of pest identified that day is extracted from the pest identification results. ,in Indicates the pest species; simultaneously obtains the daily count value of the pheromone traps. Then, a weighted fusion method was used to calculate the overall insect attraction quantity. :

[0063] ,

[0064] in, For date Pest types The total number of insects attracted, The number of pests identified from the insect-attracting images. This is the count value of the sex pheromone trap. The weight of the image recognition data is 0.6. The weight for the sex pheromone count data is 0.4. The weight setting takes into account the wide range of insect monitoring lamps and the high specificity of sex pheromone species.

[0065] Based on daily insect attraction sequence After smoothing using a five-day moving average, a pest population dynamics curve was plotted. This curve visually illustrates the changes in pest population size over time and is an important basis for judging pest dynamics. The adaptive peak detection algorithm designed in this invention identifies peak periods of pest occurrence by analyzing local peaks in the dynamics curve.

[0066] The core idea of ​​the adaptive peak detection algorithm is to use the ratio of the number of insects attracted on the current date to the average number of insects attracted in the previous period as the peak detection index. When this ratio exceeds a preset multiple, it is determined that a peak period has begun. Let the current date be... Peak period determination indicators The calculation formula is:

[0067] ,

[0068] in, For date Pest types Peak period determination indicators For date The total number of insects attracted, To calculate the average, the preceding number of days is taken as 7 days. For the first The total number of insects attracted per day. When Greater than the preset multiple threshold At that time, the date of determination Pest species The start of the peak period. In this embodiment, a preset multiple threshold is used. The setting is 2.0, which means that the peak period is determined when the number of insects attracted in a single day exceeds twice the average of the previous 7 days.

[0069] Determining the generation sequence of pests requires combining developmental time parameters from the pest's biological characteristics data. This invention uses the effective accumulated temperature method to calculate the generational development progress of pests. Effective accumulated temperature... The cumulative calculation formula is:

[0070] ,

[0071] in, From date To date The accumulated effective temperature during the period is expressed in daily degrees. For date The average daily temperature This is the temperature at which the pest begins to develop. The function ensures that effective accumulated temperature is only counted when the daily average temperature is higher than the developmental threshold temperature. A generation is considered complete when the accumulated effective temperature reaches the level required for the pest to complete one generation. For example, the rice stem borer has a developmental threshold temperature of 11.0℃, and the effective accumulated temperature for one generation is approximately 750 days. Based on this, the occurrence period of each generation of pests can be calculated.

[0072] In the implementation of this invention, the occurrence dynamic analysis module also provides various visualization functions to enable plant protection personnel to intuitively understand the occurrence dynamics of pests. The growth and decline curve graph uses date as the horizontal axis and the number of pests attracted as the vertical axis, displaying the trend of pest numbers changing over time in a line graph format, with peak periods marked by shading. The generation division graph divides the year into several generation intervals, with each interval labeled with the corresponding pest developmental generation and main insect stage. The spatial distribution map, based on data from multiple monitoring stations, displays the spatial distribution pattern of pests within the monitoring area in map form, with different colors representing different levels of occurrence. These visualization functions help plant protection personnel quickly grasp the occurrence patterns of pests, providing an intuitive basis for control decisions.

[0073] The dynamic analysis step also includes historical data comparison and analysis. The system automatically compares the current year's pest population fluctuation curve with historical data for the same period, calculating the year-on-year change rate and ranking percentile. When the number of insects attracted at a certain time in the current year is significantly higher than the historical average for the same period, the system automatically issues an early warning. In addition, the system has established a seasonal model of pest occurrence based on years of historical data, which can predict typical time nodes of pest occurrence, such as the initial appearance of overwintering adults, the peak period of the first generation, and the beginning period of the migrating generation, providing a reference for advance deployment of prevention and control preparations.

[0074] Step S4, Population Prediction Step.

[0075] The population prediction step is the core function of this invention, enabling early prediction of pest population trends. This step employs a meteorological-coupled time-series prediction network, based on historical pest population fluctuation data and meteorological condition data, to predict pest population trends for the next 7 to 14 days. Pest population dynamics are closely related to meteorological conditions; temperature affects the development rate of pests, humidity affects their survival rate, wind speed and direction affect the dispersal behavior of migratory pests, and precipitation inhibits pest activity and mating. Therefore, incorporating meteorological factors into the population prediction model is crucial for improving prediction accuracy.

[0076] The meteorological coupled time-series prediction network employs an encoder-decoder architecture, which effectively captures the temporal dependencies and the influence of meteorological conditions in pest fluctuation data. The encoder part uses a two-layer gated recurrent unit to encode the temporal features of the input pest fluctuation curve. The gated recurrent unit controls the flow of information through update and reset gate mechanisms, effectively mitigating the gradient vanishing problem during training with long-term data.

[0077] The encoder input is the past Daily pest fluctuation data sequence And the meteorological element sequence for the corresponding time period. In this embodiment, The time frame is set to 14-30 days, preferably 21 days, and the meteorological elements include daily average temperature, daily average relative humidity, daily precipitation, and daily average wind speed. The encoder first encodes the pest fluctuation sequence through a gating loop unit, generating the pest sequence hidden state. Simultaneously, the meteorological sequence is encoded through another set of gated cyclic units to generate the hidden state of the meteorological sequence. .

[0078] To effectively integrate pest sequence information and meteorological sequence information, this invention introduces an attention mechanism into the encoder. This attention mechanism dynamically allocates the degree of attention to historical time-series information based on current forecasting needs. The meteorological coupled attention weights designed in this invention... The calculation formula is:

[0079] ,

[0080] ,

[0081] in, For a moment Attention weights For a moment Attention score For a moment The hidden state of the pest sequence. For a moment The hidden state of the meteorological sequence This is the weight matrix for the latent states of pests. This is the weight matrix for the hidden meteorological states. and These are the parameter vector and bias vector for attention calculation, respectively. It is the hyperbolic tangent activation function.

[0082] Calculate the context vector based on attention weights. :

[0083] ,

[0084] in, To integrate the context vectors of pest and meteorological information. This indicates a vector concatenation operation.

[0085] The decoder is based on the context vector Gradually generating the future The predicted population size for each day. In this embodiment, The prediction period is set to 7-14 days, preferably 14 days. The decoder uses an autoregressive approach for prediction, meaning that each prediction step takes the previous step's prediction result and the context vector as input, and outputs the current prediction value. The decoder's prediction output... Indicates the future number The predicted number of insects attracted per day, of which .

[0086] The population prediction network was trained using a mean squared error loss function, and the training data was derived from historical monitoring records. This invention collected pest monitoring data and meteorological data from 50 monitoring stations across 20 provinces in China over five consecutive years, constructing a training dataset containing approximately 1.8 million records. After training and testing, the population prediction network achieved an accuracy rate of 85.2% in predicting population trends for the next 7 days and 78.6% in predicting trends for the next 14 days.

[0087] In addition to outputting the predicted insect attractant quantity, the population prediction results also output the predicted trend level. The trend level is divided into five levels: Level 1 indicates a declining population trend, Level 2 indicates a stable population, Level 3 indicates a slight increase in the population, Level 4 indicates a moderate increase in the population, and Level 5 indicates a risk of population outbreak. The trend level is determined based on the ratio of the peak insect attractant quantity during the prediction period to the historical average for the same period.

[0088] In the implementation of this invention, the performance of the meteorological coupled time-series forecasting network largely depends on the quality and diversity of the training data. To improve the model's generalization ability, this invention introduces a transfer learning strategy during network training. First, pre-training is performed on a comprehensive national dataset to learn the general patterns of pest population dynamics. Then, fine-tuning training is conducted for specific regions and specific pest species, enabling the model to capture regionally specific patterns. This two-stage training strategy allows the model to achieve good predictive results even at new monitoring sites where data is relatively scarce.

[0089] The population prediction module also provides a function for quantifying prediction uncertainty. In addition to outputting the predicted point values, the system simultaneously outputs the prediction confidence intervals. The confidence intervals are calculated using an ensemble learning method, training multiple prediction models with identical structures but different initialization parameters. The mean of the prediction results from each model is used as the final prediction value, and the standard deviation of the prediction results is used to estimate the prediction uncertainty. This uncertainty quantification allows plant protection personnel to understand the reliability of the predictions and adopt more prudent decision-making strategies under high uncertainty conditions.

[0090] Furthermore, the population prediction module implements an adaptive model update mechanism. As new monitoring data accumulates, the system periodically performs incremental training and parameter updates on the prediction model. When a significant deviation occurs between actual observations and model predictions, the system automatically detects and analyzes the causes of the deviation. Possible causes include abnormal weather conditions, sudden influx of migratory pests, and invasion of new diseases and pests. For different causes, the system takes corresponding model correction measures, such as adjusting the weights of meteorological factors and introducing migration warning signals as additional input. This adaptive update mechanism ensures that the prediction model maintains consistently high accuracy.

[0091] Step S5, Migration Early Warning Step.

[0092] The migration warning step is an innovative function of this invention for monitoring and warning of migratory pests. my country's agricultural production faces threats from several major migratory pests, including the fall armyworm, rice planthopper, armyworm, and cotton bollworm. These pests have the ability to migrate long distances and cross regions with air currents, making it difficult to provide early warnings using traditional fixed-point monitoring methods. This invention integrates insect radar monitoring data and upper-air meteorological data to achieve early warnings of the migration paths and landing areas of migratory pests.

[0093] The prerequisite for implementing the migration early warning procedure is that pest biological characteristic data indicates that the target pest has migratory characteristics. In the pest biological characteristic knowledge base, each pest is labeled with information such as whether it has migratory characteristics, its typical migration altitude range, and the migration season. For example, the fall armyworm is labeled as having migratory characteristics, with a typical migration altitude of 300 to 1500 meters and a migration season from April to October; the rice planthopper is labeled as having migratory characteristics, with a typical migration altitude of 500 to 2000 meters and a migration season from May to September.

[0094] When the system detects that a target pest exhibits migratory characteristics, the migration early warning step extracts high-altitude insect flight parameters from radar insect monitoring data. These parameters include four elements: flight altitude, flight speed, flight direction, and radar cross-section. Flight altitude is calculated using the radar's elevation angle and slant range; flight speed and flight direction are extracted using a target tracking algorithm based on continuous detection data; and radar cross-section reflects the size characteristics of the insect target and can be used to distinguish insect groups of different body sizes.

[0095] Migration trajectory inversion is a core algorithmic step in migration early warning. Because insects are significantly affected by wind fields during high-altitude flight, the flight direction and speed detected by radar are the superposition of the insect's own flight capabilities and the environmental wind field. The migration trajectory inversion algorithm designed in this invention calculates the actual displacement trajectory of the insect relative to the ground by combining radar detection parameters with high-altitude meteorological wind field data.

[0096] Assuming the insects detected by radar are at a certain time... The position is The detected flight speed was Flight direction is The wind speed provided by the upper-air meteorological data is The wind direction is Then the actual horizontal displacement velocity component of the insect relative to the ground The calculation formula is:

[0097] ,

[0098] ,

[0099] in, For the eastward velocity component, For the northward velocity component, The flight speed of insects detected by radar. The direction of flight detected by radar is increased clockwise from true north (0 degrees). For high-altitude wind speeds, The wind direction is at high altitude.

[0100] Based on the aforementioned velocity components, the insect's displacement trajectory can be calculated through time integration. From time... At the time displacement trajectory The calculation formula is:

[0101] ,

[0102] ,

[0103] in, This represents the cumulative eastward displacement. This represents the cumulative northward displacement. and They are time points The eastward and northward velocity components.

[0104] In practical calculations, numerical integration is used to discretize the trajectory. Let the time step be... Then from time Departure process Position after one time step The calculation formula is:

[0105] ,

[0106] ,

[0107] in, For the first The position coordinates of the step, These are the coordinates of the position in the previous step. and The velocity component of the previous moment. The time step is 10 minutes.

[0108] By combining migration trajectory inversion results with geographic information data, the migration paths of migratory pests and predicted landing areas can be determined. Predicting landing areas requires considering factors such as the physiological limits of insect flight time, the impact of changing weather conditions on flight, and the influence of topography on airflow. This invention divides the endpoint region of the migration trajectory into a grid, calculates the pest landing probability for each grid, and uses the region with the highest probability as the predicted landing area.

[0109] Migration early warning information includes four elements: warning level, migration time window, migration path, and predicted landing area. The warning level is divided into four levels: Level 1 is of concern, Level 2 is a warning, Level 3 is an alert, and Level 4 is an emergency. The determination of the warning level comprehensively considers factors such as the predicted peak insect population, the area of ​​agricultural production covered by the migration trajectory, and the susceptibility of local crops during their growth period.

[0110] In the implementation of this invention, the migration trajectory inversion algorithm needs to handle various complex situations. First, the upper-air wind field varies significantly with altitude, and wind speed and direction may differ significantly at different altitude levels. Therefore, it is necessary to select meteorological data for the corresponding altitude level based on the insect's actual flight altitude. In this embodiment, the upper-air meteorological data comes from radiosonde data or numerical weather prediction products released by meteorological departments, providing wind field data for every 100 meters of altitude from the ground to 3000 meters. Second, the insect's flight altitude is not constant but adjusts with changes in meteorological conditions and terrain. The migration trajectory inversion algorithm of this invention dynamically updates the insect's flight altitude parameters at each calculation time step to obtain a more accurate trajectory estimate.

[0111] The migration early warning module also integrates a function for tracing the source of migratory pests. When a local monitoring station detects a sudden surge in migratory pests, the system automatically initiates backward trajectory calculation, using the current wind conditions to infer the possible takeoff area of ​​the pests. Source tracing information is crucial for understanding the cross-regional migration patterns of pests and coordinating regional joint prevention and control efforts. In this embodiment, the backward trajectory calculation time range is set to 72 hours, allowing for the tracing of possible source areas within three days prior to the pests' takeoff.

[0112] The final step in the migration early warning process is to feed back the population prediction results to the occurrence dynamics analysis step to dynamically adjust the peak period determination parameters. When the migration early warning indicates a large-scale pest migration, the peak period determination multiple threshold needs to be lowered accordingly. This allows for more timely detection of population surges caused by immigration. This feedback mechanism enables the monitoring system of this invention to form a closed loop, dynamically optimizing monitoring strategies based on early warning information.

[0113] In the implementation of this invention, the dissemination of migratory pest warning information follows a tiered response principle. Level 1 (Attention Level) warnings are sent via internal system messages to relevant monitoring personnel, reminding them to strengthen daily monitoring. Level 2 (Alert Level) warnings are sent via SMS and mobile application push notifications to county-level plant protection departments, suggesting the conduct of field surveys. Level 3 (Warning Level) warnings simultaneously notify municipal-level plant protection departments and relevant agricultural authorities, activating emergency control plans. Level 4 (Emergency Level) warnings are reported to provincial-level authorities, coordinating cross-regional joint prevention and control resources. This tiered response mechanism ensures that warning information can be quickly transmitted to decision-makers at the appropriate levels, enabling timely and effective responses to migratory pests.

[0114] Reference Figure 2 This invention provides an intelligent monitoring and migration early warning system for farmland pests, which is used to perform the steps described in the above-mentioned method embodiments. In one embodiment of this invention, the system includes a multi-source data fusion module 1, a pest identification module 2, an occurrence dynamic analysis module 3, a population prediction module 4, and a migration early warning module 5.

[0115] The multi-source data fusion module 1 is used to acquire insect-attracting image data collected by intelligent insect monitoring lamps, insect count data collected by pheromone traps, meteorological monitoring data collected by meteorological monitoring stations, radar insect monitoring data collected by insect radar, and insect biological characteristic data from an insect biological characteristic knowledge base. It performs spatiotemporal alignment processing on the above data to generate a fused dataset. The multi-source data fusion module 1 is connected to the intelligent insect monitoring lamps, pheromone traps, meteorological monitoring stations, and insect radar via wired or wireless communication, supporting multiple communication protocols including Modbus, MQTT, and HTTP. For a detailed description of the spatiotemporal alignment processing function of the multi-source data fusion module 1, please refer to the description of the multi-source data fusion steps in the method embodiment.

[0116] The pest identification module 2 is used to input the insect-attracting image data from the fused dataset output by the multi-source data fusion module 1 into a multi-scale feature fusion target detection network, and output pest identification results including pest species, pest quantity, and identification confidence. The pest identification module 2 is deployed with a trained multi-scale feature fusion target detection network model, which supports the identification and classification of more than 100 major agricultural pests. For details on the network structure and algorithm of the pest identification module 2, please refer to the description of the pest identification steps in the method embodiment. In this embodiment, the pest identification module 2 can be deployed on an edge computing device or a cloud server. When deployed on an edge computing device equipped with a GPU, the identification processing time for a single image does not exceed 100 milliseconds.

[0117] The occurrence dynamic analysis module 3 is used to generate a daily insect attraction sequence, plot insect population fluctuation curves, identify peak periods, and determine the generation of pests based on the insect identification results output by the insect identification module 2 and the insect count data provided by the multi-source data fusion module 1. The occurrence dynamic analysis module 3 integrates an adaptive peak period detection algorithm and an effective accumulated temperature calculation function; its specific implementation is described in the description of the occurrence dynamic analysis steps in the method embodiment. The insect population fluctuation curves and peak period generation information output by the occurrence dynamic analysis module 3 are transmitted to the population prediction module 4.

[0118] The population prediction module 4 is used to predict the population trend of pests within a preset time window based on the pest population dynamics analysis module 3's output pest population fluctuation curves, peak periods, and generations, as well as the meteorological monitoring data provided by the multi-source data fusion module 1. The population prediction module 4 is equipped with a meteorological coupled time-series prediction network model; the structure and training method of this model are described in the population prediction steps of the method embodiment. The population prediction results output by the population prediction module 4 include the predicted insect attraction sequence and trend level.

[0119] The migration early warning module 5 extracts high-altitude insect flight parameters from radar insect monitoring data provided by the multi-source data fusion module 1, combines it with high-altitude meteorological data to perform migration trajectory inversion, and generates migration early warning information. The migration early warning module 5 integrates a migration trajectory inversion algorithm and a landing area prediction function; its specific implementation is described in the migration early warning steps of the method embodiment. The migration early warning module 5 also feeds back the population prediction results output by the population prediction module 4 to the occurrence dynamic analysis module 3 to dynamically adjust the peak period determination parameters, forming a closed-loop optimization mechanism for the system.

[0120] The data flow relationships between the five modules are as follows: Multi-source data fusion module 1 receives raw data from external monitoring equipment, processes it, and outputs the fused dataset; Pest identification module 2 receives image data from the fused dataset and outputs pest identification results; Occurrence dynamic analysis module 3 receives pest identification results and counting data, and outputs growth curves and peak generation information; Population prediction module 4 receives growth curves, peak generation information, and meteorological data, and outputs population prediction results; Migration early warning module 5 receives radar data and population prediction results, outputs migration early warning information, and feeds back adjustment signals to occurrence dynamic analysis module 3.

[0121] The system of this invention can be deployed on an agricultural Internet of Things platform, and push pest monitoring information, population prediction results and migration early warning information to agricultural production managers through mobile applications or web pages, supporting scientific pest control decisions.

[0122] In a specific deployment embodiment of the present invention, the system adopts a three-layer architecture: a perception layer, an edge layer, and a cloud layer. The perception layer includes data acquisition devices such as intelligent insect monitoring lamps, pheromone traps, weather monitoring stations, and insect radars, responsible for collecting raw monitoring data. The edge layer is deployed on edge computing gateway devices in the field, realizing data aggregation, preprocessing, and real-time pest identification functions. The edge gateway is equipped with an ARM processor and a neural network acceleration chip, enabling it to complete most of the computing tasks locally. The cloud layer is deployed on a data center server, undertaking computationally intensive tasks such as dynamic analysis, population prediction, and migration early warning, while also providing data storage, model training, and user services. This layered architecture ensures real-time system response while fully utilizing the powerful computing resources of the cloud.

[0123] In terms of the system's communication architecture, the perception layer devices and edge gateways utilize LoRa or NB-IoT wireless communication technologies, which feature low power consumption, long range, and low cost, making them suitable for deployment needs in agricultural scenarios. The edge gateways connect to the cloud via 4G or 5G mobile networks or wired broadband networks, using the MQTT protocol for message transmission to ensure reliable and real-time data transmission. The system supports a breakpoint resumption mechanism; when the network connection is interrupted, the edge gateway temporarily stores the data in local storage and automatically resumes transmission once the network is restored.

[0124] Regarding the system's user interface, this invention provides two access methods: a mobile application and a web-based management platform. The mobile application is designed for frontline plant protection personnel and growers, offering functions such as monitoring data query, early warning information push notifications, and prevention and control recommendations. It also supports offline data caching to cope with unstable network conditions. The web-based management platform is designed for plant protection department managers, providing functions such as multi-site data aggregation, regional analysis report generation, and historical data statistical analysis. It also supports data export and report printing. Both clients adopt a responsive design, adapting to display devices of different sizes.

[0125] Regarding system data security, this invention employs multiple security protection measures. Data transmission utilizes the TLS encryption protocol to prevent data theft or tampering during transmission. User authentication employs a token-based authentication mechanism, supporting multi-factor authentication to enhance account security. Data storage uses encrypted storage and access control, ensuring that users with different permissions can only access data within their authorized scope. The system performs regular data backups to ensure rapid recovery in case of unforeseen circumstances.

[0126] The system of this invention has achieved excellent technical results in practical applications. Pilot applications in 20 provinces across the country show that the system achieves a comprehensive identification accuracy of 91.5% for major agricultural pests, an accuracy of 84.3% for predicting population trends within 7 days, and an average lead time of 48 hours for migration warnings. Compared with traditional manual monitoring methods, the system improves monitoring efficiency by approximately 10 times, effectively reducing the workload of plant protection personnel and enhancing the timeliness and scientific rigor of pest monitoring and early warning.

[0127] The embodiments of the present invention are not limited to the specific embodiments described above. Those skilled in the art can make various equivalent changes or substitutions based on the technical solutions of the present invention, and all such changes or substitutions should be included within the protection scope of the present invention.

Claims

1. A method for intelligent monitoring and early warning of agricultural pest migration, characterized in that, include: The multi-source data fusion step involves acquiring insect-attracting image data collected by intelligent insect monitoring lamps, pest count data collected by pheromone traps, meteorological monitoring data collected by meteorological monitoring stations, radar insect monitoring data collected by insect radar, and pest biological characteristic data from a pest biological characteristic knowledge base. The insect-attracting image data, the pest count data, the meteorological monitoring data, and the radar insect monitoring data are then spatiotemporally aligned to generate a fused dataset. The pest identification step involves inputting the insect-attracting image data from the fused dataset into a multi-scale feature fusion target detection network. The multi-scale feature fusion target detection network extracts shallow texture features and deep semantic features from the insect-attracting image data, generates a pest detection feature map through cross-layer feature fusion, and outputs a pest identification result based on the pest detection feature map. The pest identification result includes pest species, pest quantity, and identification confidence. The dynamic analysis step involves generating a daily insect attraction sequence based on the pest identification results and the pest count data, plotting a pest growth curve based on the daily insect attraction sequence, performing peak detection on the pest growth curve to identify peak periods, and determining the pest generation by combining the developmental history parameters in the pest biological characteristic data. The population prediction step involves inputting the pest population fluctuation curve, the peak period, the generation, and the meteorological monitoring data into a meteorological coupled time-series prediction network. The meteorological coupled time-series prediction network predicts the pest population trend within a preset time window based on the correlation between historical population dynamics and meteorological conditions, and outputs the population prediction result. In the migration early warning step, when the biological characteristic data of the pests indicates that the target pests have migratory characteristics, the flight parameters of high-altitude insects are extracted from the radar insect monitoring data, and the migration trajectory is inverted by combining it with the high-altitude meteorological data to determine the migration path and predicted landing area of ​​the migratory pests, generate migration early warning information, and feed the population prediction results back to the occurrence dynamic analysis step to dynamically adjust the peak period determination parameters.

2. The intelligent monitoring and migration early warning method for farmland pests according to claim 1, characterized in that, In the pest identification step, when the identification confidence level is lower than a preset confidence threshold, the corresponding insect-attracting image data is marked as a sample to be reviewed. The confidence threshold ranges from 0.85 to 0.

95.

3. The intelligent monitoring and migration early warning method for farmland pests according to claim 1, characterized in that, In the dynamic analysis step, the peak detection adopts an adaptive threshold method. When the number of insects attracted in a single day of the pest fluctuation curve exceeds a preset multiple of the average value of the previous preset number of days, it is determined to be the start of the peak period. The preset number of days is 5 to 10 days, and the preset multiple is 1.5 to 3.0 times.

4. The intelligent monitoring and migration early warning method for farmland pests according to claim 1, characterized in that, In the population prediction step, the preset time window is 7 to 14 days, and the input of the meteorological coupled time series prediction network includes the pest growth and decline curves of the preset historical days and the temperature, humidity, precipitation and wind speed data of the corresponding time period, the preset historical days being 14 to 30 days.

5. The intelligent monitoring and migration early warning method for farmland pests according to claim 1, characterized in that, The multi-scale feature fusion target detection network includes a feature extraction backbone network, a multi-scale feature pyramid network, and a detection head network. The feature extraction backbone network outputs a first-scale feature map, a second-scale feature map, and a third-scale feature map. The multi-scale feature pyramid network performs bidirectional feature fusion on the first-scale feature map, the second-scale feature map, and the third-scale feature map in a top-down and bottom-up manner. The detection head network outputs pest bounding boxes and class probabilities based on the fused multi-scale feature maps.

6. The intelligent monitoring and migration early warning method for farmland pests according to claim 1, characterized in that, The meteorological coupled time-series prediction network includes an encoder and a decoder. The encoder uses a gated recurrent unit to encode the temporal features of the pest growth curve and uses an attention mechanism to fuse the meteorological monitoring data to generate a context vector. The decoder generates the population size prediction value for future time steps step by step based on the context vector.

7. The intelligent monitoring and migration early warning method for farmland pests according to claim 1, characterized in that, In the migration early warning step, the flight parameters of the high-altitude insects include flight altitude, flight speed, flight direction and radar cross section. The migration trajectory inversion is based on the flight parameters of the high-altitude insects and the wind field data in the high-altitude meteorological data to calculate the actual displacement trajectory of the insects relative to the ground.

8. The intelligent monitoring and migration early warning method for farmland pests according to claim 1, characterized in that, In the multi-source data fusion step, the spatiotemporal alignment processing includes: sorting the insect-attracting image data according to the collection timestamp, resampling the meteorological monitoring data at a preset time interval to align with the time granularity of the insect-attracting image data, and performing coordinate transformation on the radar insect monitoring data to align with the geographical coordinates of the monitoring station, wherein the preset time interval is 5 to 30 minutes.

9. The intelligent monitoring and migration early warning method for farmland pests according to claim 1, characterized in that, The pest biological characteristic data includes the pest's developmental starting temperature, effective accumulated temperature, phototactic wavelength response characteristics, migration altitude range, and migration season. The migration early warning information includes the early warning level, migration time window, migration path, and predicted landing area. The early warning level is determined based on the predicted peak insect population in the population prediction results and the coverage area of ​​the migration trajectory.

10. An intelligent monitoring and migration early warning system for farmland pests, used to implement the intelligent monitoring and migration early warning method for farmland pests as described in any one of claims 19, characterized in that, include: The multi-source data fusion module is used to acquire insect-attracting image data collected by intelligent insect monitoring lamps, pest count data collected by pheromone traps, meteorological monitoring data collected by meteorological monitoring stations, radar insect monitoring data collected by insect radar, and pest biological characteristic data in the pest biological characteristic knowledge base. The module performs spatiotemporal alignment processing on the insect-attracting image data, the pest count data, the meteorological monitoring data, and the radar insect monitoring data to generate a fused dataset. The pest identification module is used to input the insect-attracting image data in the fused dataset into a multi-scale feature fusion target detection network, and output pest identification results including pest species, pest quantity and identification confidence. The dynamic analysis module is used to generate a daily insect attraction sequence based on the pest identification results and the pest count data, draw the pest growth and decline curve, identify the peak period and determine the generation of pests. The population prediction module is used to predict the population trend of pests within a preset time window based on the pest population fluctuation curve, the peak period, the generation, and the meteorological monitoring data. The migration early warning module is used to extract high-altitude insect flight parameters from the radar insect monitoring data, combine them with high-altitude meteorological data to perform migration trajectory inversion, generate migration early warning information, and feed back the population prediction results to the occurrence dynamic analysis module to dynamically adjust the peak period determination parameters.

Images