Crop pest monitoring method and system based on artificial intelligence

Abstract

The invention discloses a crop pest monitoring method and system based on artificial intelligence, and the method comprises the steps: constructing a crop pest recognition model, and recognizing the types and number of pests; the method comprises the following steps: acquiring crop infrared thermal imaging image data of multi-view fusion, and identifying insect pest types and quantity of unshielded areas in a crop monitoring area; identifying a thermal anomaly region, and determining the number of insect pests in a shielded region in the crop monitoring region; constructing an insect pest type identification model, and determining the insect pest type of the shielded area in the crop monitoring area; calibrating the number and type data of insect pests in the crop monitoring area; constructing an insect pest prediction model, and judging a future insect pest diffusion trend; and evaluating the effectiveness of the pest control strategy, and adjusting the pest control strategy and implementing the advance of the pest control strategy. According to the invention, comprehensive monitoring, accurate identification and scientific prevention and control of insect pests are realized, the problems of monitoring blind areas and prevention and control lag are effectively solved, and the efficiency and reliability of insect pest prevention and control are significantly improved.

Description

Technical Field

[0001] The present invention relates to the field of agricultural management, and in particular to an artificial intelligence-based crop pest monitoring method and system. Background Art

[0002] As global agriculture develops towards intelligence and precision, pest monitoring and control has become an important factor affecting crop yield and quality. However, current pest monitoring technology still has many problems and is difficult to meet the needs of practical applications. Traditional monitoring methods mainly rely on optical imaging or manual observation from a single perspective. This method has obvious limitations in complex farmland environments, especially in the case of dense crop planting. The overlapping occlusion of leaves and stems leads to a large monitoring blind area, and the pest activities in the obscured area cannot be captured, which seriously limits the integrity and accuracy of the data. In addition, the image information collected from a single perspective has a limited coverage range, which makes it difficult to fully reflect the distribution of pests in the crop monitoring area, resulting in the lack of pest data in some areas, affecting the effectiveness of subsequent pest control. At the same time, traditional pest monitoring technology mainly relies on static data analysis, lacks the capture and analysis of the dynamic activity characteristics of pests, and cannot reflect the key dynamic characteristics of pests such as movement trajectory, crawling speed, and activity cycle. This static monitoring method is difficult to scientifically judge the spread trend of pests, resulting in delayed pest control. In addition, in large-scale farmland monitoring, the use of multiple devices for data collection will also cause data redundancy and inconsistency problems, especially in overlapping or adjacent areas, where the pest data collected by multiple devices are repeated, affecting the accuracy of pest quantity statistics. In addition, the inconsistency of coordinate systems between different devices also makes it difficult to integrate data, lacks a unified calibration mechanism, and increases the complexity of data processing. On the other hand, most of the existing pest control measures are passive responses, lack effective prediction of pest spread trends, and cannot be accurately intervened before pest outbreaks, missing the best prevention and control opportunities. At the same time, the effectiveness evaluation standards of prevention and control measures are not unified, and it is difficult to quantify the inhibitory effect of prevention and control strategies on the number, density and spread trends of pests, resulting in the inability to optimize and adjust the prevention and control strategies in a timely manner according to the actual pest dynamics, affecting the timeliness and accuracy of pest control. Therefore, there is an urgent need for a new method that can comprehensively and accurately monitor pest data, dynamically analyze pest spread trends, and realize scientific prediction and prevention and control decisions, so as to improve the efficiency and effectiveness of pest control and promote the development of smart agriculture. Summary of the invention

[0003] The present invention aims to solve the problems existing in the above-mentioned prior art and provide a crop pest monitoring method and system based on artificial intelligence.

[0004] The first embodiment of the present invention provides a crop pest monitoring method based on artificial intelligence, which mainly includes:

[0005] Acquire infrared thermal imaging images of crops, build crop pest recognition models, identify the types and quantities of pests in the crop monitoring area, and determine the density and activity areas of pest distribution;

[0006] By setting up infrared thermal imaging monitoring equipment at different locations in the crop monitoring area, multi-view fused infrared thermal imaging image data of crops can be obtained to identify the types and quantities of pests in the unobstructed areas of the crop monitoring area;

[0007] Based on the continuous time frame infrared thermal imaging images obtained by the thermal imaging equipment, the thermal anomaly areas are identified, the density level of pest activities in different thermal anomaly areas is determined, and the statistical results of the number of pests in the blocked areas in the crop monitoring area are obtained;

[0008] According to the shape factor, temperature fluctuation frequency, spatial dynamic characteristics, and periodic characteristics of the thermal anomaly area, a pest type identification model is constructed to determine the pest type in the obscured area of ​​the crop monitoring area;

[0009] Based on the spatial coordinates of the pest locations in the unobstructed area and the obstructed area, determine whether there are overlapping or adjacent pest locations in the unobstructed area and the obstructed area, calibrate the pest quantity and type data in the crop monitoring area, and formulate pest control strategies;

[0010] Based on the pest quantity, density and type data in the crop monitoring area, a pest prediction model is constructed to determine the future pest spread trend, and pest control strategies are used to carry out pest control in the crop monitoring area in advance;

[0011] Based on the effectiveness of pest control strategies and the lead time of pest control, the accuracy of the pest prediction model in assisting prevention and control decisions is judged, and the pest control strategies and the lead time of implementing the pest control strategies are adjusted.

[0012] Furthermore, the acquisition of crop infrared thermal imaging images, the construction of crop pest identification models, the identification of pest types and quantities in the crop monitoring area, and the determination of the density and activity area of ​​pest distribution include:

[0013] The infrared thermal imaging images of crops in the crop detection area are obtained by farmland infrared monitoring equipment, and the thermal imaging images and their corresponding time and location information are stored in the crop monitoring database; according to the infrared thermal imaging images of crops, the data annotation tools are used to annotate the pest types, pest quantities and thermal anomaly positions in the thermal imaging images to obtain labeled infrared thermal imaging datasets of crops; the labeled datasets are expanded by data enhancement methods, and the infrared thermal imaging images of crops are cleaned by image preprocessing technology to obtain data-enhanced infrared thermal imaging image datasets of crops. The data enhancement methods include rotation, scaling, translation and mirroring, and the image preprocessing technologies include normalization, denoising and histogram equalization; according to the infrared thermal imaging image datasets of crops after data enhancement, convolutional neural networks are used for model training to build a crop pest recognition model; according to the infrared thermal imaging images of crops in the crop monitoring area obtained in real time, the crop pest recognition model is used to identify the types and quantities of pests in the crop monitoring area, annotate the pest locations, determine the spatial coordinates of each pest, and judge the density and activity area of ​​pest distribution.

[0014] Furthermore, the infrared thermal imaging monitoring equipment is set at different positions in the crop monitoring area to obtain multi-view fused infrared thermal imaging image data of crops, and identify the type and quantity of pests in the unobstructed area of ​​the crop monitoring area, including:

[0015] By setting up infrared thermal imaging monitoring equipment at different positions in the crop monitoring area, multi-view infrared thermal imaging image data taken from different positions of the same crop monitoring area are obtained; based on the original infrared thermal imaging image set collected from multiple positions, the feature point matching SIFT algorithm is used to perform geometric correction and coordinate alignment on the images of each view, and the multi-view thermal imaging image data aligned in the same coordinate system is obtained through pixel-by-pixel error calculation and transformation matrix optimization; the registered multi-view infrared thermal imaging images are fused through the image fusion method to obtain the infrared thermal imaging data between different viewpoints of the same area and the comprehensive thermal map information of the thermal anomaly area, and the multi-view fused infrared thermal imaging image data of crops are obtained; based on the multi-view fused infrared thermal imaging image data of crops, the crop pest recognition model is used to identify the type and quantity of pests in the unobstructed area of ​​the crop monitoring area, mark the location of the pests, determine the spatial coordinates of each pest, and judge the density of pest distribution and the activity area.

[0016] Furthermore, the method of identifying thermal anomaly areas based on the continuous time frame infrared thermal imaging images acquired by the thermal imaging device, determining the density level of pest activities in different thermal anomaly areas, and obtaining the statistical results of the number of pests in the blocked areas in the crop monitoring area includes:

[0017] According to the continuous time frame infrared thermal imaging images obtained by the thermal imaging device, the original infrared images are denoised and enhanced by preprocessing methods to obtain time series infrared thermal imaging images; the temperature fluctuation frequency and amplitude of each pixel in the infrared thermal imaging image over time are obtained by the fast Fourier transform algorithm, and the areas where the temperature fluctuation frequency is greater than the preset frequency value and the temperature fluctuation amplitude is greater than the preset amplitude threshold are identified, and the distribution of the areas where the temperature fluctuation abnormal points are located is determined to obtain the time-frequency distribution map of the thermal abnormal area; by extracting the temperature extreme points and the spatial coordinates of the thermal abnormal area in the thermal map, combined with the threshold segmentation method and the regional growing algorithm, The thermal map is binarized, and the image areas with temperatures higher than the set first temperature threshold are marked as thermal anomaly areas, and the independent contours and boundary coordinates of each thermal anomaly area are obtained; the thermal anomaly areas in the binary image are clustered according to the density through the K-means clustering algorithm, and the center point, area size and density distribution of the thermal anomaly areas are determined to obtain independent thermal anomaly distribution cluster data; based on the thermal anomaly distribution cluster data, the density level of pest activities in different thermal anomaly areas is determined by counting the size, density and number of hotspots of the thermal anomaly areas, and the statistical results of the number of pests in the blocked areas in the crop monitoring area are obtained.

[0018] Furthermore, the method of constructing a pest type recognition model based on the shape factor, temperature fluctuation frequency, spatial dynamic characteristics, and periodic characteristics of the thermal anomaly area to determine the pest type of the obscured area in the crop monitoring area includes:

[0019] According to the extracted data of thermal anomaly areas, the moving path and speed of the thermal anomaly areas in continuous time frames are calculated to obtain the moving speed and motion trajectory data of the area, and the spatial dynamic characteristics of pest activities, including crawling speed and movement mode, are obtained; according to the duration of the extracted dynamic heat signal, the window sliding average method is used to identify the time change length and period of the heat signal, and the duration of the temperature signal in each thermal anomaly area is obtained to determine the periodic characteristics of different pest activities, among which the shape factor Among them, P i is the perimeter of the potential temperature anomaly area, A i is the area of ​​the potential temperature anomaly area; according to the shape factor, temperature fluctuation frequency, spatial dynamic characteristics, and periodic characteristics of the thermal anomaly area, a heat map dynamic feature vector dataset is constructed and saved in the pest activity database; according to the heat map dynamic feature vector dataset with pest types marked in the pest activity database, a random forest algorithm is used for model training to build a pest type recognition model; according to the real-time monitored heat map dynamic feature vector, the pest type recognition model is used to determine the pest type of each thermal anomaly area and determine the pest type of the blocked area in the crop monitoring area.

[0020] Furthermore, based on the spatial coordinates of the pest positions in the unobstructed area and the obstructed area, it is determined whether there are overlapping or adjacent pest positions in the unobstructed area and the obstructed area, and the pest quantity and type data in the crop monitoring area are calibrated, and pest control strategies are formulated, including:

[0021] Through spatial coordinate mapping, the spatial coordinates of the pest positions in the unobstructed area and the obstructed area are mapped to the unified coordinate system of the same monitoring area; the position of each coordinate point in the unified coordinate system is calculated by using affine transformation or perspective transformation, and the pest spatial coordinates and bounding box mapped to the same coordinate system are obtained; according to the pest position spatial coordinate data mapped to the same coordinate system, the IoU intersection-over-union method is used to determine whether there are overlapping or adjacent pest positions in the unobstructed area and the obstructed area, and determine the overlapping distribution of pests in the overlapping or adjacent areas, including the number of pests, the position of the bounding box and the spatial relationship; according to the pest type recognition results of the unobstructed area and the obstructed area, the cosine similarity calculation method is used to calculate the dynamic characteristics of the heat map. The similarity of feature vector data is used to determine whether the types of pests in the overlapping area or the adjacent area are consistent, and the pest type labels of the overlapping area or the adjacent area are determined; if the types of pests in the overlapping area or the adjacent area are consistent, the number of pests in the overlapping area or the adjacent area is calibrated by the deduplication method, including removing the pest records with a small number of pests in the overlapping area or the adjacent area, retaining only the pest records with a large number of pests, and recording the number and type data of pests in the crop monitoring area; according to the number and type data of pests in the crop monitoring area, a pest control strategy is formulated, and the pest control strategy is implemented by drones to control pests in the crop monitoring area. The pest control strategy includes the use of pesticide spraying, biological control or physical control methods to carry out early control of the pest area.

[0022] Furthermore, the pest prediction model is constructed based on the pest quantity, density and type data in the crop monitoring area to determine the future pest spread trend, and pest control strategies are used to carry out pest control in the crop monitoring area in advance, including:

[0023] Continuously monitor the number, density and type data of insect pests in the crop monitoring area, use long short-term memory network for model training, build an insect pest prediction model, predict the number, density and type data of insect pests in the crop monitoring area in the future preset time period, and judge the future trend of insect pest spread; if it is predicted that the growth rate of the number of insect pests in the crop monitoring area in the future preset time period is greater than the preset rate threshold, or the density is greater than the preset degree threshold, then use the insect pest control strategy to carry out insect pest control in the crop monitoring area in advance.

[0024] Further, based on the effectiveness of the pest control strategy and the advance time of the pest control opportunity, judging the auxiliary accuracy of the pest prediction model for the control decision, adjusting the pest control strategy and the advance time of implementing the pest control strategy, includes:

[0025] Obtain changes in pest density, distribution and type after pest control, and judge the inhibitory effect of pest control strategies on the number and spread of pests, and the effectiveness of pest control strategies based on the comparison of pest data after pest control with the prediction results before control; if the effectiveness is lower than the preset effectiveness threshold, adjust the pest control strategy; use the pest prediction model and the trigger data of the pest control strategy to compare the predicted pest growth rate with the actual pest reduction rate to evaluate the rationality and lead time of the pest control timing, judge the auxiliary accuracy of the pest prediction model for control decision-making, and obtain evaluation data on the lead time and timeliness of the pest control strategy. If the accuracy is lower than the preset accuracy threshold, adjust the lead time of implementing the pest control strategy.

[0026] The second aspect of the present invention provides an artificial intelligence-based crop pest monitoring system, which mainly includes:

[0027] The crop pest analysis module is used to obtain infrared thermal imaging images of crops, build crop pest recognition models, identify the types and quantities of pests in the crop monitoring area, and determine the density and activity areas of pest distribution;

[0028] The unobstructed area pest identification module is used to obtain multi-view fused crop infrared thermal imaging image data through infrared thermal imaging monitoring equipment set at different positions in the crop monitoring area, and identify the type and quantity of pests in the unobstructed area of ​​the crop monitoring area;

[0029] The module for analyzing the number of pests in the obscured area is used to identify the thermal anomaly area based on the continuous time frame infrared thermal imaging images obtained by the thermal imaging device, determine the density level of pest activities in different thermal anomaly areas, and obtain the statistical results of the number of pests in the obscured area in the crop monitoring area;

[0030] The module for identifying the type of pests in the obscured area is used to construct an identification model for the type of pests according to the shape factor, temperature fluctuation frequency, spatial dynamic characteristics, and periodic characteristics of the thermal anomaly area, and to determine the type of pests in the obscured area in the crop monitoring area;

[0031] The pest data calibration module is used to determine whether there are overlapping or adjacent pest locations in the unobstructed area and the obstructed area based on the pest location spatial coordinates in the unobstructed area and the obstructed area, calibrate the pest quantity and type data in the crop monitoring area, and formulate pest control strategies;

[0032] The pest spread trend prediction module is used to build a pest prediction model based on the pest quantity, density and type data in the crop monitoring area, determine the future pest spread trend, and use pest control strategies to carry out pest control in the crop monitoring area in advance;

[0033] The pest control strategy adjustment module is used to judge the auxiliary accuracy of the pest prediction model for control decision-making based on the effectiveness of the pest control strategy and the lead time of the pest control opportunity, adjust the pest control strategy and the lead time of implementing the pest control strategy.

[0034] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0035] The present invention provides a crop pest monitoring method and system based on artificial intelligence. The present invention improves the comprehensiveness of monitoring in unobstructed areas through multi-view fusion technology, and uses continuous time frame thermal imaging data to identify thermal anomaly areas and analyze the density level and statistical number of pests in obstructed areas, effectively making up for the monitoring blind spots in obstructed areas. The present invention combines the shape factor, temperature fluctuation frequency, dynamic characteristics and periodic characteristics of the pest area to further construct a pest type recognition model, and realizes accurate judgment of the pest type in the obstructed area. The present invention calibrates the spatial coordinates of the unobstructed area and the obstructed area to eliminate data overlap and redundancy, thereby improving the accuracy and consistency of monitoring data. The present invention uses pest quantity, density and type data to construct a pest prediction model, judge the pest spread trend, and conduct precise intervention in advance in combination with pest control strategies, effectively avoiding the further spread of pests and reducing crop losses. The present invention also improves the timeliness and accuracy of pest control by evaluating the effectiveness of the prevention and control strategy and the rationality of the prevention and control timing, dynamically adjusting the prevention and control strategy and advance amount, realizing efficient closed-loop management of pest monitoring, prediction and prevention, and promoting the intelligent and precise development of smart agricultural management. The present invention realizes comprehensive monitoring, accurate identification and scientific prevention and control of pests through multi-perspective fusion, dynamic thermal anomaly analysis and pest prediction model, effectively solves the problems of monitoring blind spots, data redundancy and prevention and control lag, significantly improves the efficiency and reliability of pest control, and provides technical support for the development of smart agriculture. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A flowchart of a crop pest monitoring method based on artificial intelligence of the present invention;

[0037] Figure 2 The present invention provides a process for crop pest monitoring method based on artificial intelligence;

[0038] Figure 3 A schematic diagram of a crop pest monitoring system based on artificial intelligence of the present invention; DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0040] like Figure 1-2 In this embodiment, a crop pest monitoring method based on artificial intelligence may specifically include:

[0041] Step S101, acquiring infrared thermal imaging images of crops, constructing a crop pest recognition model, identifying the types and quantities of pests in the crop monitoring area, and determining the density of pest distribution and the activity area.

[0042] The infrared thermal imaging images of crops in the crop detection area are obtained by the farmland infrared monitoring equipment, and the thermal imaging images and their corresponding time and location information are stored in the crop monitoring database. According to the infrared thermal imaging images of crops, the data annotation tools are used to annotate the pest types, pest quantities and thermal anomaly positions in the thermal imaging images to obtain the labeled crop infrared thermal imaging dataset. The labeled dataset is expanded by the data enhancement method, and the infrared thermal imaging images of crops are cleaned by the image preprocessing technology to obtain the data-enhanced crop infrared thermal imaging image dataset. The data enhancement methods include rotation, scaling, translation and mirroring, and the image preprocessing techniques include normalization, denoising and histogram equalization. According to the data-enhanced crop infrared thermal imaging image dataset, the convolutional neural network is used for model training to build a crop pest recognition model. According to the infrared thermal imaging images of crops in the crop monitoring area obtained in real time, the crop pest recognition model is used to identify the types and quantities of pests in the crop monitoring area, annotate the pest locations, determine the spatial coordinates of each pest, and judge the density and activity area of ​​pest distribution.

[0043] Exemplarily, infrared thermal imaging images of the crop detection area are obtained through farmland infrared monitoring equipment. In a 5-hectare farmland area, the equipment captures a total of 500 infrared thermal imaging images, and stores these images and their shooting time 08:00-12:00 on June 1, 2020 and the corresponding geographic location information coordinates X=123.456, Y=456.789 in the crop monitoring database. Based on these infrared thermal imaging images, the data annotation tool is used to manually annotate the types of pests in the image, including rice planthoppers and rice leaf rollers, the number of pests, including 150 rice planthoppers and 50 rice leaf rollers, and the abnormal thermal positions of pests, including X=123.46 to 123.50, Y=456.78 to 456.80, to obtain a labeled crop infrared thermal imaging dataset containing pest annotation information. Through data enhancement methods, such as rotating the original image by 90 degrees, scaling it to 0.8 times, translating it left and right by 10 pixels, and mirroring it vertically, 500 annotated images were expanded to 2,000 data-enhanced images. Image preprocessing techniques, such as normalizing the image pixel value range to between 0 and 1, denoising to filter out high-frequency noise, and improving the image brightness distribution through histogram equalization, were used to obtain a quality-optimized data-enhanced crop infrared thermal imaging image dataset. A convolutional neural network was used to train on this dataset. After 20 rounds of training, the model recognition accuracy reached 95.2%. When the infrared thermal imaging image of the crop monitoring area is obtained in real time, the image of the area is analyzed using the trained crop pest recognition model, and the pest types are identified as 100 rice planthoppers and 40 rice leaf folders. The spatial position coordinates of each pest are marked, and the center coordinates of the rice planthopper hotspot are: X=123.47, Y=456.79. The density of the pest distribution is also counted, and the average number of pests per square meter is 2.8. It is judged that the rice planthoppers are mainly concentrated in the southern area of ​​field A, while the rice leaf folder is more dispersed.

[0044] Step S102, by setting up infrared thermal imaging monitoring equipment at different positions in the crop monitoring area, multi-view fused infrared thermal imaging image data of crops is obtained to identify the types and quantities of pests in the unobstructed areas of the crop monitoring area.

[0045] By setting up infrared thermal imaging monitoring equipment at different positions in the crop monitoring area, multi-view infrared thermal imaging image data taken from different positions in the same crop monitoring area are obtained. According to the original infrared thermal imaging image set collected from multiple positions, the feature point matching SIFT algorithm is used to perform geometric correction and coordinate alignment on each view image. Through pixel-by-pixel error calculation and transformation matrix optimization, the aligned multi-view thermal imaging image data in the same coordinate system are obtained. The registered multi-view infrared thermal imaging images are fused by the image fusion method to obtain the infrared thermal imaging data between different viewpoints in the same area and the comprehensive thermal map information of the thermal anomaly area, and obtain the multi-view fused infrared thermal imaging image data of crops. According to the multi-view fused infrared thermal imaging image data of crops, the crop pest recognition model is used to identify the type and quantity of pests in the unobstructed area of ​​the crop monitoring area, mark the location of pests, determine the spatial coordinates of each pest, and judge the density and activity area of ​​pest distribution.

[0046] For example, infrared monitoring equipment is set up in four different directions of the crop monitoring area, such as east, south, west, and north, to obtain infrared thermal imaging image data of the same farmland area of ​​2 hectares. The equipment in each direction takes 100 infrared images, forming a total of 400 multi-view thermal imaging image sets. Based on these multi-directional image data, the SIFT feature point matching algorithm is used to extract key feature points in the image, such as extracting 500 feature points from each image on average. Then, the error threshold is set to 0.5 pixels through pixel-by-pixel error calculation, and geometric correction and coordinate alignment are performed on each perspective image. After transformation matrix optimization, all images are finally aligned to the same coordinate system to obtain spatially consistent registered multi-view infrared thermal imaging image data. Through weighted average fusion, the aligned images are fused, and the temperature data of different viewing angles and the information of thermal anomaly areas are combined to obtain a comprehensive thermal map of the farmland area, showing the pest activity area with hot spot temperatures ranging from 30.5℃ to 33.2℃. The fused thermal map is analyzed using the crop pest recognition model to identify the types of pests in the unobstructed area, such as 150 rice planthoppers and 60 rice leaf folders, and the spatial coordinates of these pests are marked, such as the coordinates of the center point of the rice planthopper hot spot area are X=123.47, Y=456.79. At the same time, it is determined that rice planthoppers are mainly distributed in the southwest area of ​​the farmland, with a density of 5 per square meter, while the activity area of ​​rice leaf folders is relatively scattered, with an average of 2 per square meter. The peak activity area of ​​rice planthoppers is determined to be in the southwest field.

[0047] Step S103, identifying the thermal anomaly area based on the continuous time frame infrared thermal imaging image acquired by the thermal imaging device, determining the density level of pest activities in different thermal anomaly areas, and obtaining the statistical result of the number of pests in the shielded area in the crop monitoring area.

[0048] According to the continuous time frame infrared thermal imaging images obtained by the thermal imaging device, the original infrared images are de-noised and enhanced by the preprocessing method to obtain the time series infrared thermal imaging images. The fluctuation frequency and amplitude of the temperature value of each pixel in the infrared thermal imaging image over time are obtained by the fast Fourier transform algorithm, and the areas where the temperature fluctuation frequency is greater than the preset frequency value and the temperature fluctuation amplitude is greater than the preset amplitude threshold are identified, and the distribution of the areas where the temperature fluctuation abnormal points are located is determined to obtain the time-frequency distribution diagram of the thermal abnormal area. By extracting the spatial coordinates of the temperature extreme points and the thermal abnormal area in the thermal map, combining the threshold segmentation method and the regional growing algorithm, the thermal map is binarized, and the image area with a temperature higher than the set first temperature threshold is marked as the thermal abnormal area, and the independent contour and boundary coordinates of each thermal abnormal area are obtained. The thermal abnormal area in the binary image is clustered according to the density by the K-means clustering algorithm, and the center point, area size and density distribution of the thermal abnormal area are determined to obtain independent thermal abnormality distribution cluster data. According to the thermal anomaly distribution cluster data, the density level of pest activities in different thermal anomaly areas is determined by counting the size, density and number of hot spots of the thermal anomaly areas, and the statistical results of the number of pests in the obscured areas in the crop monitoring area are obtained.

[0049] For example, based on the continuous time frame infrared thermal imaging images obtained by the thermal imaging device, in a crop monitoring area of ​​1 hectare, the device continuously shoots 300 frames of infrared thermal imaging images at a rate of 1 frame per second to obtain the temperature data of each pixel changing over time. The original thermal imaging image is subjected to Gaussian filtering noise reduction and contrast stretching enhancement through image preprocessing methods to remove environmental noise and enhance the details of the hot spot area. The fast Fourier transform algorithm is applied to the preprocessed time series thermal map data to analyze the fluctuation frequency and amplitude of the temperature value of each pixel over time, extract the area with a temperature fluctuation frequency greater than the preset value of 0.1Hz and a temperature fluctuation amplitude greater than 2°C, identify the distribution of temperature fluctuation abnormal points in space, and obtain the time-frequency distribution map of the thermal abnormal area covering the entire monitoring area. In the thermal map, the temperature extreme points with a temperature exceeding 32°C and their corresponding spatial coordinates X=100.5, Y=200.3 are extracted. All pixel areas with a temperature above 30°C are marked as thermal anomaly areas by combining the threshold segmentation method. Then, the marked area is expanded by the regional growing algorithm to generate independent contours and boundary coordinates of each thermal anomaly area. For example, the boundary coordinate range of a thermal anomaly area is X=100.0 to 105.0, Y=200.0 to 205.0. By applying the K-means clustering algorithm to the binary image data, these thermal anomaly areas are divided into three clusters according to their density. The average density of cluster 1 is 10 thermal anomaly points per square meter, cluster 2 is 5 thermal anomaly points per square meter, and cluster 3 is 2 thermal anomaly points per square meter. The center point of each cluster is determined, such as the center point coordinates of cluster 1 are X=102.5, Y=202.5, and the area size of each cluster is calculated, such as the area of ​​cluster 1 is 5 square meters. According to the cluster data of the thermal anomaly area distribution cluster after clustering, the size, density and number of hotspots of the thermal anomaly area of ​​each cluster were counted to determine the density level of pests in the obscured area. For example, the pest density in Cluster 1 was high risk, with about 3 pests per square meter and a total of 15 pests, while Clusters 2 and 3 were medium and low risk areas, respectively, with a total of 12 pests. Finally, the statistical result of the pests in the entire obscured area was 27.

[0050] Step S104, constructing an insect pest type recognition model based on the shape factor, temperature fluctuation frequency, spatial dynamic characteristics, and periodic characteristics of the thermal anomaly area to determine the insect pest type of the obscured area in the crop monitoring area.

[0051] According to the extracted data of thermal anomaly areas, the moving path and speed of the thermal anomaly areas in continuous time frames are calculated to obtain the moving speed and motion trajectory data of the area, and the spatial dynamic characteristics of pest activities, including crawling speed and movement mode, are obtained. According to the duration of the extracted dynamic heat signal, the window sliding average method is used to identify the time change length and period of the heat signal, and the duration of the temperature signal of each thermal anomaly area is obtained to determine the periodic characteristics of different pest activities, among which the shape factor Among them, P i is the perimeter of the potential temperature anomaly area, A i is the area of ​​the potential temperature anomaly region. According to the shape factor, temperature fluctuation frequency, spatial dynamic characteristics, and periodic characteristics of the thermal anomaly region, a heat map dynamic feature vector dataset is constructed and saved to the pest activity database. According to the heat map dynamic feature vector dataset with pest types marked in the pest activity database, the random forest algorithm is used for model training to construct a pest type recognition model. According to the real-time monitored heat map dynamic feature vector, the pest type recognition model is used to determine the pest type of each thermal anomaly region and determine the pest type of the obscured area in the crop monitoring area.

[0052] For example, based on the extracted thermal anomaly area data, in a 2-hectare farmland monitoring area, the moving path and speed of the thermal anomaly area are calculated through the analysis of infrared thermal imaging data of continuous time frames, and the moving speed of one of the thermal anomaly areas is obtained to be 0.15 m / s. A corresponding motion trajectory is generated, and the trajectory shows that its moving direction is a linear crawling mode, moving from the coordinate point X=120.5, Y=200.8 to X=123.0, Y=203.5. According to the time change of the dynamic thermal signal, the window sliding average method is used to calculate the duration of the thermal signal, and it is determined that the duration of the temperature signal in the thermal anomaly area is 180 seconds, and the periodic characteristics are manifested as a temperature fluctuation peak every 40 seconds. According to the calculation formula of the shape factor Calculate the shape factor φ of the thermal anomaly area i , where the perimeter P i =10.5m, area A i =8.7m 2, the shape factor of the thermal anomaly area is calculated to be 1.12, indicating that the area is an irregular but nearly circular thermal anomaly shape. The shape factor of the thermal anomaly area, the temperature fluctuation frequency of 0.025Hz, the movement speed and the periodic characteristics are combined to construct a dynamic feature vector [0.15, 180, 1.12, 0.025] and save it to the pest activity database. The random forest algorithm is used for model training based on the labeled heat map dynamic feature vector dataset in the pest activity database to build a pest type recognition model with a training accuracy of 93.5%. When the dynamic feature vector [0.15, 180, 1.12, 0.025] of the real-time monitored heat map is input into the pest type recognition model, the model determines that the thermal anomaly area belongs to the rice planthopper activity area, and determines that the pest type in the thermal anomaly area is rice leafhopper, and marks that the pest is located in the coordinate area X = 120.5 to X = 123.0, Y = 200.8 to Y = 203.5. Therefore, the pest type of the blocked area in the crop monitoring area is rice leafhopper.

[0053] Step S105, based on the spatial coordinates of the pest positions in the unobstructed area and the obstructed area, determine whether there are overlapping or adjacent pest positions in the unobstructed area and the obstructed area, calibrate the pest quantity and type data in the crop monitoring area, and formulate a pest control strategy.

[0054] Through spatial coordinate mapping, the spatial coordinates of the pest positions in the unobstructed area and the obstructed area are mapped to the unified coordinate system of the same monitoring area. Affine transformation or perspective transformation is used to calculate the position of each coordinate point in the unified coordinate system to obtain the pest spatial coordinates and bounding box mapped to the same coordinate system. According to the pest position spatial coordinate data mapped to the same coordinate system, the IoU intersection-over-union method is used to determine whether there are overlapping or adjacent pest positions in the unobstructed area and the obstructed area, and to determine the overlapping distribution of pests in the overlapping or adjacent areas, including the number of pests, the position of the bounding box and the spatial relationship. According to the pest type recognition results of the unobstructed area and the obstructed area, the cosine similarity calculation method is used to calculate the similarity of the dynamic feature vector data of the heat map, to determine whether the pest types in the overlapping area or the adjacent area are consistent, and to determine the pest type labeling of the overlapping area or the adjacent area. If the types of pests in the overlapping or adjacent areas are the same, the number of pests in the overlapping or adjacent areas is calibrated by deduplication, including removing pest records with small numbers of pests in the overlapping or adjacent areas, retaining only pest records with large numbers of pests, and recording the number and type data of pests in the crop monitoring area. According to the number and type data of pests in the crop monitoring area, a pest control strategy is formulated, and the pest control strategy is implemented by drones to control pests in the crop monitoring area. The pest control strategy includes the use of pesticide spraying, biological control or physical control methods to prevent and control pests in the pest area in advance.

[0055] Exemplarily, through spatial coordinate mapping, the spatial coordinates of the pest positions in the unobstructed area and the obscured area are mapped to a unified coordinate system in the same monitoring area. For example, the pest position coordinates in the unobstructed area are (X=120.5, Y=200.3) and (X=122.0, Y=202.5), and the pest coordinates in the obscured area are (X=120.6, Y=200.4) and (X=122.1, Y=202.6). All coordinate points are transformed into the same coordinate system through affine transformation to obtain unified coordinate points of the pests. According to the mapped pest location data, the Intersection over Union (IoU) method is used to calculate the degree of overlap of the pest bounding box. If the bounding box overlap rate (IoU) of the unobstructed area and the obstructed area is 0.75, it is judged that there is an overlap in the distribution of pests in these two areas, and the number of pests in the overlapping area is determined to be 7. The bounding box position covers the coordinates X = 120.4 to 122.2, Y = 200.2 to 202.8, and the spatial position relationship is recorded. According to the results of pest type recognition, the unobstructed area identifies the pest type as rice planthoppers, and the obstructed area identifies the pest type as rice planthoppers. The similarity of the dynamic feature vectors of the two heat maps is calculated by the cosine similarity calculation method, and the dynamic feature vector similarity reaches 97%, which is consistent with the pest type in the overlapping area. Since there are duplicate records of the number of pests in the overlapping area, the records with a larger number of pests are retained after deduplication. For example, in the overlapping area, the number of pests recorded in the unobstructed area is 5, and the number of pests recorded in the obstructed area is 4. After removing a small number of overlapping records, the total number of pests in the overlapping area is determined to be 5, and the total number of pests in the crop monitoring area is recorded as 23, of which rice planthoppers account for the majority. According to the number and type data of pests, pest control strategies are formulated, and pesticide spraying methods are selected to control rice planthoppers in high-density areas. UAVs are used to evenly spray pesticides in the hot spots of rice planthopper activity in the coordinate area X=120.4 to 122.2, Y=200.2 to 202.8 to complete the pest control task in the crop monitoring area.

[0056] Step S106, constructing a pest prediction model based on the pest quantity, density and type data in the crop monitoring area, judging the future pest spread trend, and using pest control strategies to carry out pest control in the crop monitoring area in advance.

[0057] Continuously monitor the number, density and type of pests in the crop monitoring area, use long short-term memory network for model training, build a pest prediction model, predict the number, density and type of pests in the crop monitoring area in the future preset time period, and judge the future pest spread trend. If the predicted growth rate of the number of pests in the crop monitoring area in the future preset time period is greater than the preset rate threshold, or the density is greater than the preset degree threshold, then use the pest control strategy to carry out pest control in the crop monitoring area in advance.

[0058] For example, by continuously monitoring the pest data in the crop monitoring area, in a 3-hectare farmland, the infrared thermal imaging equipment obtains the number, density and type data of pests every hour in one day. The number of recorded pests increased from 50 at 8 am to 150 at 2 pm, and the density increased from an average of 2 per square meter to 5. The pest types are mainly rice planthoppers and rice leaf rollers. These time series data are input into the long short-term memory network for model training, and the model is used to predict the changing trend of pests in the next 12 hours. The results show that the number of pests will increase to 200 in the next 6 hours, with a growth rate of 8.3% per hour, exceeding the preset rate threshold of 5%. At the same time, the density of pests is expected to reach 6 per square meter in the next 6 hours, exceeding the preset density threshold of 4 per square meter. According to the prediction results, it is judged that the spread trend of pests is more serious and may further aggravate the damage. Then the pest control strategy is adopted to carry out early prevention and control in the crop monitoring area. UAVs are selected to spray pesticides in areas with high pest density, such as coordinates X=120.5 to 125.5, Y=200.0 to 205.0, and precise prevention and control is carried out in the hot distribution areas of rice planthoppers and rice leaf folders to prevent the further spread of pests and reduce the threat of pests to crops.

[0059] Step S107, based on the effectiveness of the pest control strategy and the advance time of the pest control opportunity, determine the auxiliary accuracy of the pest prediction model for control decision-making, adjust the pest control strategy and the advance time of implementing the pest control strategy.

[0060] Obtain changes in pest density, distribution, and type after pest control. Compare pest data after pest control with prediction results before control to determine the inhibitory effect of pest control strategies on pest quantity and spread, and determine the effectiveness of pest control strategies. If the effectiveness is lower than the preset effectiveness threshold, adjust the pest control strategy. Use the pest prediction model and the trigger data of the pest control strategy to compare the predicted pest growth rate with the actual pest reduction rate to evaluate the rationality and lead time of pest control timing, determine the auxiliary accuracy of the pest prediction model for control decision-making, and obtain evaluation data on the lead time and timeliness of the pest control strategy. If the accuracy is lower than the preset accuracy threshold, adjust the lead time of the pest control strategy.

[0061] For example, in a 2-hectare farmland monitoring area, after the pesticide spraying control strategy was implemented by drones, the pest control effect was evaluated, and the density, distribution and type data of pests after control were obtained. The number of pests decreased from 200 before control to 120, and the average density decreased from 6 per square meter to 3.6. The pest types were still mainly rice planthoppers and rice leaf rollers. The data after control was compared with the model prediction results before control. The prediction results showed that the number of pests was reduced to 90, and the density was reduced to 3 per square meter. It was judged that the inhibition rate of the control strategy on the number of pests was 40%, which was lower than the preset effectiveness threshold of 50%, indicating that the current control strategy was not effective enough and the pesticide dosage or control coverage needed to be adjusted. At the same time, the pest prediction model and the data triggered by the control strategy were used to compare the predicted pest growth rate of 8.3% per hour with the actual pest reduction rate of 5% per hour to evaluate the rationality of the control timing. The results showed that the lead time for control was 2 hours, but because the actual reduction rate was lower than expected, the pest spread inhibition effect was not optimal. By analyzing the lead time and timeliness of pest control strategies, it is determined that the current pest prediction model's auxiliary accuracy for control decisions is 85%, which is lower than the preset accuracy threshold of 90%. Therefore, the control triggering time needs to be advanced by 1 hour to ensure that control measures are implemented earlier to improve the control effect. The final optimized strategy will be implemented in the next round of pest prediction and control.

[0062] like Figure 3 In this embodiment, a crop pest monitoring system based on artificial intelligence may specifically include:

[0063] The crop pest analysis module is used to obtain infrared thermal imaging images of crops, build crop pest recognition models, identify the types and quantities of pests in the crop monitoring area, and determine the density and activity area of ​​pest distribution.

[0064] The unobstructed area pest identification module is used to obtain multi-view fused crop infrared thermal imaging image data through infrared thermal imaging monitoring equipment set at different positions in the crop monitoring area, and identify the type and quantity of pests in the unobstructed area in the crop monitoring area.

[0065] The module for analyzing the number of pests in obscured areas is used to identify thermal anomaly areas based on the continuous time frame infrared thermal imaging images obtained by the thermal imaging device, determine the density level of pest activities in different thermal anomaly areas, and obtain statistical results of the number of pests in obscured areas in the crop monitoring area.

[0066] The module for identifying pest types in obscured areas is used to construct a pest type identification model based on the shape factor, temperature fluctuation frequency, spatial dynamic characteristics, and periodic characteristics of the thermal anomaly area, and to determine the pest type in the obscured area in the crop monitoring area.

[0067] The pest data calibration module is used to determine whether there are overlapping or adjacent pest positions in the unobstructed area and the obstructed area based on the spatial coordinates of the pest positions in the unobstructed area and the obstructed area, calibrate the pest quantity and type data in the crop monitoring area, and formulate pest control strategies.

[0068] The pest spread trend prediction module is used to build a pest prediction model based on the pest quantity, density and type data in the crop monitoring area, determine the future pest spread trend, and use pest control strategies to carry out pest control in the crop monitoring area in advance.

[0069] The pest control strategy adjustment module is used to judge the auxiliary accuracy of the pest prediction model for control decision-making based on the effectiveness of the pest control strategy and the lead time of the pest control opportunity, adjust the pest control strategy and the lead time of implementing the pest control strategy.

[0070] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the concept of the present application. For example, the above features are replaced with the technical features with similar functions disclosed in the present application (but not limited to) to form a technical solution.

Claims

1. A crop pest monitoring method based on artificial intelligence, characterized in that: The method comprises: Acquire infrared thermal imaging images of crops, build crop pest recognition models, identify the types and quantities of pests in the crop monitoring area, and determine the density and activity areas of pest distribution; By setting up infrared thermal imaging monitoring equipment at different locations in the crop monitoring area, multi-view fused infrared thermal imaging image data of crops can be obtained to identify the types and quantities of pests in the unobstructed areas of the crop monitoring area; Based on the continuous time frame infrared thermal imaging images obtained by the thermal imaging equipment, the thermal anomaly areas are identified, the density level of pest activities in different thermal anomaly areas is determined, and the statistical results of the number of pests in the blocked areas in the crop monitoring area are obtained; According to the shape factor, temperature fluctuation frequency, spatial dynamic characteristics, and periodic characteristics of the thermal anomaly area, a pest type identification model is constructed to determine the pest type in the obscured area of ​​the crop monitoring area; Based on the spatial coordinates of the pest locations in the unobstructed area and the obstructed area, determine whether there are overlapping or adjacent pest locations in the unobstructed area and the obstructed area, calibrate the pest quantity and type data in the crop monitoring area, and formulate pest control strategies; Based on the pest quantity, density and type data in the crop monitoring area, a pest prediction model is constructed to determine the future pest spread trend, and pest control strategies are used to carry out pest control in the crop monitoring area in advance; Based on the effectiveness of pest control strategies and the lead time of pest control, the accuracy of the pest prediction model in assisting prevention and control decisions is judged, and the pest control strategies and the lead time of implementing the pest control strategies are adjusted.

2. The method according to claim 1, wherein: The method of acquiring infrared thermal imaging images of crops, constructing a crop pest identification model, identifying the types and quantities of pests in the crop monitoring area, and determining the density and activity area of ​​pest distribution includes: The infrared thermal imaging images of crops in the crop detection area are obtained by farmland infrared monitoring equipment, and the thermal imaging images and their corresponding time and location information are stored in the crop monitoring database; according to the infrared thermal imaging images of crops, the data annotation tools are used to annotate the pest types, pest quantities and thermal anomaly positions in the thermal imaging images to obtain labeled infrared thermal imaging datasets of crops; the labeled datasets are expanded by data enhancement methods, and the infrared thermal imaging images of crops are cleaned by image preprocessing technology to obtain data-enhanced infrared thermal imaging image datasets of crops. The data enhancement methods include rotation, scaling, translation and mirroring, and the image preprocessing technologies include normalization, denoising and histogram equalization; according to the infrared thermal imaging image datasets of crops after data enhancement, convolutional neural networks are used for model training to build a crop pest recognition model; according to the infrared thermal imaging images of crops in the crop monitoring area obtained in real time, the crop pest recognition model is used to identify the types and quantities of pests in the crop monitoring area, annotate the pest locations, determine the spatial coordinates of each pest, and judge the density and activity area of ​​pest distribution.

3. The method according to claim 1, wherein: The infrared thermal imaging monitoring equipment is set at different positions in the crop monitoring area to obtain multi-view fused infrared thermal imaging image data of crops, and identify the types and quantities of pests in the unobstructed areas of the crop monitoring area, including: By setting up infrared thermal imaging monitoring equipment at different positions in the crop monitoring area, multi-view infrared thermal imaging image data taken from different positions of the same crop monitoring area are obtained; based on the original infrared thermal imaging image set collected from multiple positions, the feature point matching SIFT algorithm is used to perform geometric correction and coordinate alignment on the images of each view, and the multi-view thermal imaging image data aligned in the same coordinate system is obtained through pixel-by-pixel error calculation and transformation matrix optimization; the registered multi-view infrared thermal imaging images are fused through the image fusion method to obtain the infrared thermal imaging data between different viewpoints of the same area and the comprehensive thermal map information of the thermal anomaly area, and the multi-view fused infrared thermal imaging image data of crops are obtained; based on the multi-view fused infrared thermal imaging image data of crops, the crop pest recognition model is used to identify the type and quantity of pests in the unobstructed area of ​​the crop monitoring area, mark the location of the pests, determine the spatial coordinates of each pest, and judge the density of pest distribution and the activity area.

4. The method according to claim 1, wherein: The method of identifying thermal anomaly areas based on the continuous time frame infrared thermal imaging images obtained by the thermal imaging device, determining the density level of pest activities in different thermal anomaly areas, and obtaining the statistical results of the number of pests in the blocked areas in the crop monitoring area includes: According to the continuous time frame infrared thermal imaging images obtained by the thermal imaging device, the original infrared images are subjected to noise reduction and enhancement processing by a preprocessing method to obtain a time series infrared thermal imaging image; the fluctuation frequency and amplitude of the temperature value of each pixel in the infrared thermal imaging image over time are obtained by a fast Fourier transform algorithm, and the areas where the temperature fluctuation frequency is greater than a preset frequency value and the temperature fluctuation amplitude is greater than a preset amplitude threshold are identified, and the distribution of the areas where the temperature fluctuation abnormal points are located is determined to obtain a time-frequency distribution diagram of the thermal abnormal area; by extracting the temperature extreme points and the spatial coordinates of the thermal abnormal area in the thermal map, the thermal map is binarized by combining the threshold segmentation method and the regional growing algorithm, and the image area with a temperature higher than the set first temperature threshold is marked as the thermal abnormal area, and the independent contour and boundary coordinates of each thermal abnormal area are obtained; the thermal abnormal area in the binary image is clustered according to the density by the K-means clustering algorithm, and the center point, area size and density distribution of the thermal abnormal area are determined to obtain independent thermal abnormality distribution cluster data; According to the thermal anomaly distribution cluster data, the density level of pest activities in different thermal anomaly areas is determined by counting the size, density and number of hot spots of the thermal anomaly areas, and the statistical results of the number of pests in the obscured areas in the crop monitoring area are obtained.

5. The method according to claim 1, wherein: The method of constructing a pest type recognition model based on the shape factor, temperature fluctuation frequency, spatial dynamic characteristics, and periodic characteristics of the thermal anomaly area to determine the pest type of the obscured area in the crop monitoring area includes: According to the extracted data of thermal anomaly areas, the moving path and speed of the thermal anomaly areas in continuous time frames are calculated to obtain the moving speed and motion trajectory data of the area, and the spatial dynamic characteristics of pest activities, including crawling speed and movement mode, are obtained; according to the duration of the extracted dynamic heat signal, the window sliding average method is used to identify the time change length and period of the heat signal, and the duration of the temperature signal in each thermal anomaly area is obtained to determine the periodic characteristics of different pest activities, among which the shape factor Among them, P i is the perimeter of the potential temperature anomaly area, A i is the area of ​​the potential temperature anomaly area; according to the shape factor, temperature fluctuation frequency, spatial dynamic characteristics, and periodic characteristics of the thermal anomaly area, a heat map dynamic feature vector dataset is constructed and saved in the pest activity database; according to the heat map dynamic feature vector dataset with pest types marked in the pest activity database, a random forest algorithm is used for model training to build a pest type recognition model; according to the real-time monitored heat map dynamic feature vector, the pest type recognition model is used to determine the pest type of each thermal anomaly area and determine the pest type of the blocked area in the crop monitoring area.

6. The method according to claim 1, wherein: Based on the spatial coordinates of the pest positions in the unobstructed area and the obstructed area, it is determined whether there are overlapping or adjacent pest positions in the unobstructed area and the obstructed area, and the pest quantity and type data in the crop monitoring area are calibrated, and pest control strategies are formulated, including: Through spatial coordinate mapping, the spatial coordinates of the pest locations in the unobstructed area and the obstructed area are mapped to a unified coordinate system in the same monitoring area; Using affine transformation or perspective transformation, the position of each coordinate point in the unified coordinate system is calculated to obtain the pest space coordinates and bounding box mapped to the same coordinate system; Based on the spatial coordinate data of pest locations mapped to the same coordinate system, the IoU intersection-over-union method is used to determine whether there are overlapping or adjacent pest locations in the unoccluded area and the occluded area, and to determine the overlapping distribution of pests in the overlapping or adjacent areas, including the number of pests, the position of the bounding box, and the spatial relationship. According to the pest type identification results of the unobstructed area and the obstructed area, the cosine similarity calculation method is used to calculate the similarity of the dynamic feature vector data of the heat map, determine whether the pest types in the overlapping area or the adjacent area are consistent, and determine the pest type labeling of the overlapping area or the adjacent area; if the pest types in the overlapping area or the adjacent area are consistent, the number of pests in the overlapping area or the adjacent area is calibrated by the deduplication method, including removing the pest records with a small number of pests in the overlapping area or the adjacent area, retaining only the pest records with a large number of pests, and recording the number and type data of pests in the crop monitoring area; according to the number and type data of pests in the crop monitoring area, a pest control strategy is formulated, and the pest control strategy is implemented by drones to carry out pest control in the crop monitoring area. The pest control strategy includes the use of pesticide spraying, biological control or physical control methods to carry out early control of the pest area.

7. The method according to claim 1, wherein: The pest prediction model is constructed based on the pest quantity, density and type data in the crop monitoring area to determine the future pest spread trend, and pest control strategies are used to carry out pest control in the crop monitoring area in advance, including: Continuously monitor the number, density and type of pests in the crop monitoring area, use long short-term memory networks for model training, build a pest prediction model, predict the number, density and type of pests in the crop monitoring area within a preset time period in the future, and determine the future pest spread trend; If it is predicted that the growth rate of the number of pests in the crop monitoring area within a preset time period in the future is greater than the preset rate threshold, or the density is greater than the preset degree threshold, the pest control strategy will be used to carry out pest control in the crop monitoring area in advance.

8. The method according to claim 1, wherein: The effectiveness of the pest control strategy and the advance time of the pest control opportunity are used to judge the auxiliary accuracy of the pest prediction model for the control decision, adjust the pest control strategy and the advance time of implementing the pest control strategy, including: Obtain changes in pest density, distribution and type after pest control, and judge the inhibitory effect of pest control strategies on the number and spread of pests, and the effectiveness of pest control strategies based on the comparison of pest data after pest control with the prediction results before control; if the effectiveness is lower than the preset effectiveness threshold, adjust the pest control strategy; use the pest prediction model and the trigger data of the pest control strategy to compare the predicted pest growth rate with the actual pest reduction rate to evaluate the rationality and lead time of the pest control timing, judge the auxiliary accuracy of the pest prediction model for control decision-making, and obtain evaluation data on the lead time and timeliness of the pest control strategy. If the accuracy is lower than the preset accuracy threshold, adjust the lead time of implementing the pest control strategy.

9. An artificial intelligence-based crop pest monitoring system, executing the steps of an artificial intelligence-based crop pest monitoring method according to any one of claims 1 to 8, characterized in that: The system comprises: The crop pest analysis module is used to obtain infrared thermal imaging images of crops, build crop pest recognition models, identify the types and quantities of pests in the crop monitoring area, and determine the density and activity areas of pest distribution; The unobstructed area pest identification module is used to obtain multi-view fused crop infrared thermal imaging image data through infrared thermal imaging monitoring equipment set at different positions in the crop monitoring area, and identify the type and quantity of pests in the unobstructed area of ​​the crop monitoring area; The module for analyzing the number of insect pests in the obscured area is used to identify the thermal anomaly area based on the continuous time frame infrared thermal imaging images obtained by the thermal imaging device, determine the density level of insect pest activities in different thermal anomaly areas, and obtain the statistical results of the number of insect pests in the obscured area in the crop monitoring area; The module for identifying the type of pests in the obscured area is used to construct an identification model for the type of pests according to the shape factor, temperature fluctuation frequency, spatial dynamic characteristics, and periodic characteristics of the thermal anomaly area, and to determine the type of pests in the obscured area in the crop monitoring area; The pest data calibration module is used to determine whether there are overlapping or adjacent pest locations in the unobstructed area and the obstructed area based on the pest location spatial coordinates in the unobstructed area and the obstructed area, calibrate the pest quantity and type data in the crop monitoring area, and formulate pest control strategies; The pest spread trend prediction module is used to build a pest prediction model based on the pest quantity, density and type data in the crop monitoring area, determine the future pest spread trend, and use pest control strategies to carry out pest control in the crop monitoring area in advance; The pest control strategy adjustment module is used to judge the auxiliary accuracy of the pest prediction model for control decision-making based on the effectiveness of the pest control strategy and the lead time of the pest control opportunity, adjust the pest control strategy and the lead time of implementing the pest control strategy.

Images