Area pest early warning and prevention system and method based on visual recognition

Abstract

This invention belongs to the field of pest monitoring technology, specifically a regional pest early warning and control system and method based on visual recognition. The system includes a multispectral pest trapping and imaging module, an image noise adaptive filtering module, a pest dual-branch feature analysis module, a pest population density quantification module, a pest spread trend prediction module, a precise control instruction adaptation module, and a pest control evaluation and optimization module. This invention collects multispectral pest images and environmental parameters through trapping, denoises and fuses them, and then uses a dual-branch network to simultaneously extract morphological and spectral features to accurately identify pests. It also quantifies pest population density and, combined with environmental parameters, predicts spread trends, classifies into three levels of early warning, and generates precise control instructions. This achieves accurate pest identification, early warning, and scientific control, improving the efficiency and level of regional pest early warning and control, and is suitable for large-scale agricultural production.

Description

Technical Field

[0001] This invention relates to the field of pest monitoring technology, specifically a regional pest early warning and control system and method based on visual recognition. Background Technology

[0002] Pests are one of the core issues affecting agricultural production. Especially in large-scale planting scenarios, the rapid spread of pests can easily lead to reduced crop yields, decreased quality, and even regional agricultural losses. Therefore, achieving early and accurate identification, trend prediction, and scientific control of pests is crucial to ensuring stable agricultural production and reducing resource waste. With the development of large-scale and intelligent agriculture, traditional pest control methods that rely on manual inspections and experience-based judgments are no longer suitable for the needs of efficient supervision. Various pest early warning and control solutions based on technological means are gradually emerging.

[0003] Currently, traditional pest monitoring and control technologies rely on trapping devices combined with simple image acquisition to capture pests, and some introduce single visual recognition models to identify pest types. Control measures are then formulated based on human experience or fixed thresholds. This approach has not formed a multi-dimensional collaborative management and control system and has the following core shortcomings: First, most of them use single-spectral image acquisition, lack multispectral data fusion and adaptive noise reduction processing, are easily affected by background interference such as field weeds and leaf shadows, have insufficient ability to distinguish between small pests and similar pests, have low identification accuracy, and have not established a dynamic correlation between pest characteristics and environmental parameters. Population density quantification is simply calculated by "number of pests / area", ignoring the influence of environmental factors such as temperature and wind speed, resulting in large quantification errors and failing to truly reflect the degree of pest occurrence. Secondly, the lack of a scientific model for predicting the spread trend makes it impossible to predict the short-term spread situation in advance, resulting in delayed early warnings and missed opportunities for optimal prevention and control. Furthermore, the generation of prevention and control instructions lacks specificity and fails to accurately match the pest species, spread level, and crop type. The general adoption of uniform prevention and control methods and pesticide dosages can easily lead to pesticide waste, environmental pollution, or insufficient prevention and control intensity. There is also no mechanism for quantitatively evaluating the effectiveness of prevention and control. After long-term operation, the system's accuracy gradually declines, making it difficult to meet the long-term needs of regional pest monitoring.

[0004] Therefore, developing a regional pest early warning and control system that can integrate multispectral data to achieve accurate pest identification, combine environmental parameters to quantify population density, scientifically predict spread trends, classify early warning levels, and generate targeted control instructions has become an urgent technical problem to be solved in the field of pest supervision. Summary of the Invention

[0005] The purpose of this invention is to provide a regional pest early warning and control system and method based on visual recognition, so as to solve the technical defects mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a regional pest early warning and control system based on visual recognition, including a pest multispectral trapping imaging module, an image noise adaptive filtering module, a pest dual-branch feature analysis module, a pest population density quantification module, a pest spread trend prediction module, and a precision control instruction adaptation module. The insect multispectral trapping and imaging module is used to design trapping structures by combining the color attraction and phototaxis of insects, and simultaneously acquires insect images and environmental parameters of the monitoring area in different spectral bands; the image noise adaptive filtering module adaptively adjusts the filtering parameters according to the noise characteristics of different spectral images, removes image noise and fuses multispectral images; the insect dual-branch feature analysis module constructs a dual-branch feature analysis network to simultaneously extract the morphological and spectral features of insects, and accurately identify and distinguish the types of insects. The pest population density quantification module quantifies pest population density based on the output of the pest dual-branch feature analysis module and combined with monitoring area parameters. The pest spread trend prediction module predicts the short-term spread trend of pests based on the pest population density quantification results and pest species information, combined with environmental parameters, and outputs the pest spread warning level. The precision control instruction adaptation module generates targeted precision control instructions based on the pest spread warning level and pest species, combined with the crop type in the monitoring area.

[0007] Furthermore, the first branch of the dual-branch feature parsing network is the morphological feature parsing branch, which extracts morphological features including the outline of the pest, the number of limbs, the distribution of wing veins and the size of the body through convolutional layers and pooling layers, and uses an attention mechanism to focus on the key parts of the pest. The second branch of the dual-branch feature parsing network is the spectral feature parsing branch, which extracts the reflectance differences of insect pests in different spectral bands in the fused image and generates spectral feature vectors.

[0008] Furthermore, the specific process for quantifying pest population density is as follows: The system receives parameters from the pest dual-branch feature analysis module and the pest multispectral trapping imaging module, and then calculates the environmental correction coefficient using environmental parameters. After obtaining the environmental correction coefficient k, the pest population density is calculated using the population density quantification formula. After quantification, the pest population density ρ and various parameters in the calculation process are transmitted to the pest spread trend prediction module.

[0009] Furthermore, the specific process by which the pest spread trend prediction module predicts the short-term spread trend of pests is as follows: Based on the type of pest, the preset pest diffusion rate baseline value ro is called. The ro is different for different pests. Then, the diffusion rate baseline value is corrected in combination with environmental parameters to obtain the actual diffusion rate r of the corresponding pest. After calculating the actual spread rate r of the pest, a simplified reaction-diffusion model is used to predict the pest population density ρt after time t using the exponential growth formula, and the pest population density increment Δρ of the corresponding pest within the prediction period is calculated. After the prediction is completed, the module classifies the pest warning level according to the size of Δρ, and transmits the pest spread prediction information, including the pest warning level, ρt and Δρ, to the precision prevention and control instruction adaptation module.

[0010] Furthermore, the strategy for classifying pest warning levels is as follows: If Δρ≤0.5, the pest is considered to be spreading slowly and no emergency control is required, and it is marked as a Level 1 warning; if 0.5<Δρ≤1.5, the pest is considered to be spreading moderately and control measures need to be prepared, and it is marked as a Level 2 warning; if 1.5<Δρ, the pest is considered to be spreading rapidly and control measures need to be prepared immediately, and it is marked as a Level 3 warning.

[0011] Furthermore, after receiving information on the pest spread warning level, predicted population density, and pest type, the precision prevention and control instruction adaptation module calls upon the preset crop-pest-control scheme database to determine the control method based on the pest type. Subsequently, based on the pest spread warning level and predicted population density, it determines the control intensity and scope. At the same time, combined with the crop type in the monitoring area, it adjusts the concentration, dosage, and application time of the control agent, generates detailed control instructions, and transmits them to the terminal equipment of the management personnel.

[0012] Furthermore, the specific methods for determining prevention and control methods are as follows: For small piercing-sucking insect pests, biological control should be the priority. For large chewing insect pests, a combination of physical and chemical control methods should be used. For pests that spread rapidly and in high densities, chemical control should be the primary method, supplemented by biological control.

[0013] Furthermore, the specific methods for determining the intensity and scope of prevention and control are as follows: During a Level 1 alert, only localized control measures will be implemented in areas where pests are concentrated. During a Level II alert, routine prevention and control measures are implemented across the entire monitoring area. When a Level III warning is issued, the scope of prevention and control measures should be expanded and the frequency of prevention and control measures should be increased.

[0014] Furthermore, both the pest spread trend prediction module and the precision control instruction adaptation module are communicatively connected to the pest control evaluation and optimization module. The pest spread trend prediction module sends pest spread prediction information to the pest control evaluation and optimization module, and the precision control instruction adaptation module sends control execution feedback information to the pest control evaluation and optimization module. The pest control evaluation and optimization module evaluates the pest control effect and calculates the pest control efficiency ηc. If the pest control efficiency ηc ≥ 80%, the pest control is deemed qualified; if the pest control efficiency ηc < 80%, the pest control is deemed unqualified.

[0015] Furthermore, if pest control is unsatisfactory, the reasons will be analyzed, including incorrect pest identification, biased spread prediction, and unreasonable control instructions. The relevant module parameters will then be calibrated accordingly, as follows: If the pest identification is incorrect, update the feature template in the pest feature database; If the spread prediction is biased, adjust the baseline value of the pest spread rate and the relevant parameters of the environmental correction coefficient. If the prevention and control instructions are unreasonable, optimize the crop-pest-control program database.

[0016] The present invention also proposes a regional pest early warning and control method based on visual recognition, which includes the following steps: Step 1: Pest spectral trapping and transportation; Step 2: Adaptive purification of insect-like noise; Step 3: Deconstruction of the two-dimensional characteristics of pests; Step 4: Quantify the population density of pests; Step 5: Predicting the spread of pests; Step Six: Generating Precision Prevention and Control Instructions; Step 7: Feedback on pest control.

[0017] Compared with the prior art, the beneficial effects of the present invention are: 1. In this invention, multispectral trapping and collection, adaptive noise reduction and fusion, combined with dual-branch feature analysis are used to achieve accurate identification and species differentiation of pests. Furthermore, the invention quantifies pest population density and predicts spread trends, classifies three levels of early warning, adapts to pest and crop types, and generates precise control instructions. This solves the problems of inaccurate identification, delayed early warning, and blind control in traditional methods, improves the efficiency of pest supervision and the accuracy of control, and reduces resource waste.

[0018] 2. In this invention, the pest control assessment and optimization module accurately calculates the assessment effect of control efficiency, calibrates relevant parameters to address issues such as identification errors, prediction deviations, and unreasonable instructions, further improving the system's operational reliability and the sustainability of control effects, and adapting to the long-term monitoring needs of regional pests. Attached Figure Description

[0019] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a system block diagram of Embodiment 1 and Embodiment 2 of the present invention; Figure 2 This is a flowchart of the method in Embodiment 3 of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1: Refer to Figure 1 As shown in the figure, the regional pest early warning and control system based on visual recognition proposed in this embodiment consists of a pest multispectral trapping imaging module, an image noise adaptive filtering module, a pest dual-branch feature analysis module, a pest population density quantification module, a pest spread trend prediction module, and a precision control command adaptation module.

[0022] The insect multispectral trapping and imaging module is used to design trapping structures by combining insect color attraction and phototaxis, and simultaneously acquires insect images and environmental parameters of the monitoring area in different spectral bands. It solves the problems of blurry images, overlapping insects, and missing environmental parameters in existing acquisition modules, which lead to large errors in subsequent analysis, and provides high-quality, multi-dimensional raw data for subsequent modules.

[0023] Specifically, the insect pest multispectral trapping imaging module is equipped with an adjustable multispectral camera, a gradient yellow trapping board (adapted to the color attraction of most agricultural pests, wavelength 520-580nm), and a low-power LED attractant lamp (wavelength 365nm, to avoid strong light causing pest overlap); it is installed in the monitoring area at preset intervals (for example, one per 50 mu in open fields and one per 100 square meters in greenhouses) to form a grid-like monitoring layout.

[0024] Furthermore, once the multispectral insect trapping and imaging module is activated, the trapping board attracts flying insects through color attraction, while the LED light attracts nocturnal insects. Simultaneously, the multispectral camera automatically switches between three core spectral bands (visible light 400–760 nm, near-infrared 760–1100 nm, and ultraviolet 300–400 nm). Preferably, an image is acquired every 15 minutes. During the acquisition process, environmental parameters such as temperature T, wind speed V, and light intensity L at the monitoring point are simultaneously collected by the built-in sensor.

[0025] After acquisition, the original images are initially standardized (uniform resolution of 1920×1080, saved as PNG format), and the multispectral images (each band is imaged separately) are packaged with environmental parameters and transmitted in real time to the image noise adaptive filtering module via wireless transmission (5G / Wi-Fi dual mode).

[0026] The image noise adaptive filtering module receives the raw data transmitted by the insect pest multispectral trapping imaging module. Based on the noise characteristics of different spectral images (illuminance noise in visible light images, sensor noise in near-infrared images, and interference noise in ultraviolet images), it adaptively adjusts the filtering parameters to remove image noise and fuse multispectral images, thereby improving image clarity and providing high-quality image data for subsequent insect pest feature analysis. This avoids the problems of incomplete noise filtering or loss of image details caused by existing fixed parameter filtering.

[0027] Specifically, after receiving the multispectral images and environmental parameters transmitted by the insect pest multispectral trapping imaging module, the image noise adaptive filtering module first identifies the noise type of the images in the three bands. Combined with synchronously acquired parameters such as light intensity L and temperature T, it determines the source of noise (e.g., when the light intensity is below 500 lux, visible light images are prone to light noise, and when the temperature is above 30℃, near-infrared images are prone to sensor thermal noise).

[0028] Subsequently, an adaptive filtering algorithm is activated to adjust the filtering parameters for different noise types: For visible light images, an adaptive Gaussian filter is used (the filter kernel size is dynamically adjusted according to the light intensity L; when L < 500 lux, the filter kernel size is 5×5, and when L ≥ 500 lux, the filter kernel size is 3×3) to remove noise caused by uneven illumination. For near-infrared images, median filtering (with a fixed kernel size of 3×3) is used to remove sensor thermal noise; for ultraviolet images, bilateral filtering is used to preserve pest details while removing interference noise.

[0029] After filtering, the images of the three bands are pixel-fused. By weighted superposition of pixel gray values, the gray value difference between the pests and the background (weeds, soil, crop leaves) is highlighted, and a clear image after fusion is generated. Simultaneously, the fused image and the filtered environmental parameters are transmitted to the pest dual-branch feature analysis module, which effectively solves the problem of fixed parameters and poor adaptability of the existing filtering module and improves the accuracy of subsequent feature recognition.

[0030] The dual-branch feature analysis module for pests breaks through the limitations of existing YOLO series models. It constructs a dual-branch feature analysis network to simultaneously extract morphological and spectral features of pests, enabling accurate identification and species differentiation of small pests (such as thrips and aphids). This solves the technical pain points of existing feature extraction modules, which can only identify large pests, have low accuracy in complex backgrounds, and have ambiguous species differentiation.

[0031] Specifically, after receiving the fused image and environmental parameters transmitted by the image noise adaptive filtering module, the pest dual-branch feature analysis module first performs preprocessing on the fused image (image normalization and contrast enhancement), normalizing the image pixel values ​​to the [0,1] range, and enhancing the contrast between pests and the background through adaptive histogram equalization, thus solving the problem of unclear pest features caused by complex field backgrounds (such as weed cover and leaf shadows).

[0032] Subsequently, a dual-branch feature parsing network is launched: the first branch is the morphological feature parsing branch, which extracts morphological features such as the outline, number of limbs, distribution of wing veins, and body size of the pest through convolutional layers and pooling layers, and uses an attention mechanism (SMCA spatial multi-scale contextual attention) to focus on key parts of the pest and avoid background interference. The second branch is the spectral feature analysis branch, which extracts the reflectance differences of pests in different spectral bands in the fused image and generates spectral feature vectors (e.g., aphids have significantly higher reflectance in the near-infrared band than weeds, and thrips have specific reflectance in the ultraviolet band).

[0033] After the feature vectors of the two branches are fused, they are input into a preset pest feature library (containing morphological feature templates and spectral feature templates of 50 common regional pests, generated through training with more than 100,000 samples). The cosine similarity matching algorithm is used for comparison. When the similarity is ≥85%, it is determined to be the corresponding pest type. At the same time, the total number of identified pests N, the location coordinates of a single pest, and the morphological parameters (body length a, body width b, etc.) are output. If the similarity is <85%, it is marked as an unknown pest and an alert signal is triggered simultaneously.

[0034] Furthermore, after identification is completed, pest identification information such as pest type, quantity N, location coordinates, and morphological parameters are transmitted to the pest population density quantification module and the pest spread trend prediction module, effectively solving the problems of existing feature extraction modules relying on single morphological features, low identification accuracy, and inability to distinguish similar pests.

[0035] The pest population density quantification module, based on the output of the pest dual-branch feature analysis module and combined with monitoring area parameters, achieves accurate quantification of pest population density. It introduces an environmental correction coefficient to address the problems of existing quantification methods neglecting the influence of environmental factors and having large quantification errors. Specifically, the operational analysis process for quantifying pest population density is as follows: First, the system receives the pest quantity N, location coordinates, and morphological parameters transmitted from the pest dual-branch feature analysis module, and simultaneously retrieves parameters such as the monitoring area S. Then, it calculates the environmental correction coefficient using environmental parameters, as shown in the following formula: ; k: Environmental correction factor, used to correct the effects of temperature, humidity and wind speed on the spatial distribution of pests; T: Temperature at the monitoring point, acquired by the temperature sensor built into the insect multispectral trapping imaging module; To: Suitable temperature reference value, fixed at 25℃; V: Wind speed at the monitoring point, acquired by the wind speed sensor built into the insect multispectral trapping imaging module.

[0036] After calculating the environmental correction coefficient k, the pest population density is calculated using the population density quantification formula, which is as follows: ; ρ: Pest population density in the monitored area, a core indicator for judging the degree of pest occurrence; N: The total number of valid pests identified by the pest dual-branch feature analysis module, obtained after image recognition to remove overlaps and fuzzy annotations; S: The actual area of ​​the monitoring area, calculated by converting the latitude and longitude boundary coordinates of the insect multispectral trapping imaging module based on GPS positioning. k: Environmental correction factor.

[0037] It should be noted that when environmental conditions are close to the optimal value (T≈25℃, V≈0), k≈1, the correction effect is minimal, and the trapping data best reflects the true population density; when environmental conditions deviate from the optimal value, k>1, by increasing the denominator, the calculated population density is reduced, correcting the trapping deviation caused by environmental factors, which helps to avoid misjudging the degree of pest infestation.

[0038] After quantification, the pest population density ρ and various parameters in the calculation process are transmitted to the pest spread trend prediction module. The quantification accuracy is significantly improved compared with the existing technology, which solves the problem that the existing quantification methods simply calculate by "total number of insects / area" and ignore environmental factors.

[0039] The pest spread trend prediction module, based on the quantification results of pest population density and pest species information, combined with environmental parameters, enables accurate prediction of the short-term (1-7 days) spread trend of pests and outputs a spread warning level. This solves the problem that current methods can only determine the current severity of pests but cannot predict the spread trend, resulting in delayed pest warnings. Specifically, the operational analysis process for predicting the short-term spread trend of pests is as follows: First, it receives pest type information transmitted by the pest dual-branch feature analysis module, current population density ρo (the population density of the corresponding pest at the current moment) transmitted by the pest population density quantification module, and retrieves environmental parameters (temperature T, wind speed V, etc.) output by the image noise adaptive filtering module. Then, based on the pest type, the preset pest diffusion rate baseline value ro is called. It should be noted that the pest diffusion rate baseline value ro is different for different pest types, and this value is obtained by fitting historical monitoring data. Subsequently, the diffusion rate baseline value is corrected by combining environmental parameters to obtain the actual diffusion rate r of the corresponding pest. The calculation formula is as follows: ; Where r: the actual spread rate of the pest; ro: Baseline value for pest spread rate, retrieved from the system database based on the pest type; T: Temperature at the monitoring point, acquired by the sensor; To: Suitable temperature reference value, fixed at 25℃; V: Wind speed at the monitoring point, acquired by sensors.

[0040] It should be noted that higher temperatures accelerate pest metabolism, making them more active in flight and crawling, increasing their motivation to forage and migrate, and significantly accelerating their spread. The higher the wind speed, the more difficult it is for pests to spread on their own. Therefore, the formula, with positive temperature correction and negative wind speed correction, perfectly matches the actual spread behavior of pests in small farmland, which helps to ensure the authenticity and accuracy of the spread rate calculation results.

[0041] After calculating the actual spread rate r of the pest, a simplified reaction-diffusion model is used to predict the pest population density ρt after time t using the exponential growth formula, and the increase in the corresponding pest population density Δρ within the prediction period is calculated. The specific calculation formula is as follows: ; ; where ρt: the predicted insect population density after time t; ρo: The current pest population density, directly output by the pest population density quantification module; e: natural constant, fixed value 2.718; r: actual rate of pest spread; t: Predicted duration, set by the user, with a range of 1 to 7 days, and a default of 3 days; Δρ: Increment of pest population density within the predicted period.

[0042] After the prediction is completed, the module classifies the pest warning level according to the size of Δρ, and transmits the pest spread prediction information, including the pest warning level, ρt and Δρ, to the precision prevention and control instruction adaptation module. The strategy for classifying pest warning levels is as follows: If Δρ≤0.5, it is determined that the pest is spreading slowly and no emergency control is needed, and it is marked as a Level 1 warning. If 0.5 < Δρ ≤ 1.5, then the pest spread is considered moderate, and prevention and control measures should be prepared and the warning should be marked as a level 2 warning. If 1.5 < Δρ, it is determined that the pest is spreading rapidly and requires immediate control measures, and it is marked as a Level 3 warning.

[0043] The precision prevention and control instruction adaptation module generates targeted precision prevention and control instructions based on the pest spread warning level and pest type, combined with the crop type in the monitoring area. This enables targeted treatment and precise prevention and control, improving the efficiency and level of regional pest early warning and prevention and control, and is suitable for large-scale agricultural production.

[0044] Specifically, after receiving information such as the pest spread warning level, predicted population density, and pest type, the precision prevention and control instruction adaptation module calls upon a pre-set crop-pest-control scheme database (containing optimal control methods for different crops and pests, such as adaptation conditions for physical control, biological control, and chemical control) to determine the control method based on the pest type, as shown in the following example: For small piercing-sucking pests (such as aphids and thrips), biological control (such as releasing natural enemies) should be given priority. For large chewing pests (such as cabbage caterpillars and codling moths), physical control (such as trapping) combined with chemical control is used. For pests that spread rapidly and in high densities, chemical control should be the primary method, supplemented by biological control.

[0045] Subsequently, based on the pest spread warning level and the predicted population density, the intensity and scope of control measures are determined, as shown in the following example: During a Level 1 alert, only localized control measures will be implemented in areas where pests are concentrated. During a Level II alert, routine prevention and control measures are implemented across the entire monitoring area. When a Level III warning is issued, the prevention and control area should be expanded (preferably 50m beyond the monitoring area), and the frequency of prevention and control should be increased. At the same time, based on the crop type in the monitoring area (such as wheat, fruit trees, vegetables, etc.), the concentration, dosage and application time of the control agents are adjusted (e.g., avoid applying chemical agents to vegetables within 7 days before harvest), and detailed control instructions are generated (including control method, agent name, dosage, application time, control range and personnel to be implemented) and transmitted to the terminal equipment of the management personnel.

[0046] Example 2: Refer to Figure 1 As shown, the difference between this embodiment and Embodiment 1 is that in this embodiment, both the pest spread trend prediction module and the precision control command adaptation module are communicatively connected to the pest control evaluation and optimization module. The pest spread trend prediction module sends pest spread prediction information to the pest control evaluation and optimization module, and the precision control command adaptation module sends control execution feedback information to the pest control evaluation and optimization module. The pest control evaluation and optimization module evaluates the pest control effect and calculates the pest control efficiency ηc, using the following formula: ; Wherein, ρt: the predicted pest population density before prevention and control, output by the pest spread trend prediction module; ρafter: The actual pest population density measured after the control operation is completed. It is obtained by re-collecting, identifying and analyzing the pests after the control operation.

[0047] The effectiveness of pest control is judged based on the pest control efficiency ηc. If the efficiency ηc ≥ 80%, the pest control is considered effective; if ηc < 80%, the pest control is considered ineffective. If the pest control is ineffective, the reasons should be analyzed (such as incorrect pest identification, biased spread prediction, unreasonable control instructions, etc.), and the relevant module parameters should be calibrated accordingly. See the following for reference: If the pest identification is incorrect, update the feature template in the pest feature database; If the spread prediction is biased, adjust the baseline value of pest spread rate ro and related parameters of the environmental correction coefficient; If the prevention and control instructions are unreasonable, optimize the crop-pest-control program database.

[0048] Example 3: Refer to Figure 2 As shown, the difference between this embodiment and Embodiments 1 and 2 is that this embodiment proposes a regional pest early warning and control method based on visual recognition, including the following steps: Step 1: Pest Spectrum Trapping and Transportation: The insect pest multispectral trapping imaging module attracts insect pests through trapping structures, collects multispectral images and parameters such as temperature, humidity, light, wind speed, and monitoring area, and outputs them in real time after standardized processing. Step 2: Adaptive purification of insect-like noise: The image noise adaptive filtering module receives data, identifies noise types and performs targeted noise removal, fuses multispectral images and generates a clear image, and transmits the fused image and environmental parameters to the pest dual-branch feature analysis module. Step 3: Deconstructing the Two-Dimensional Characteristics of Pests: After image preprocessing, the insect pest dual-branch feature analysis module extracts the morphological and spectral features of insect pests, matches them with the insect pest feature library to determine the species, and outputs parameters such as the number of insect pests and their location coordinates. Step 4: Quantifying the population density of pests: The pest population density quantification module receives relevant parameters, calculates the pest population density, and transmits the results to the pest spread trend prediction module. Step 5: Prediction of Pest Spread Trends: The pest spread trend prediction module calculates the actual spread rate and predicts the population density and spread increment after time t, and classifies the warning into three levels accordingly. Step Six: Generating Precision Prevention and Control Instructions: The precision prevention and control instruction adaptation module combines the early warning level, pest type, etc., to determine the prevention and control methods, intensity, and scope, and generate detailed prevention and control instructions; Step 7: Pest Control Feedback The pest control assessment and optimization module calculates the control efficiency and determines whether the pest control is up to standard.

[0049] The working principle of this invention is as follows: Using visual recognition and multispectral technology as its core, the multispectral insect trapping and imaging module collects three-band insect images and environmental parameters such as temperature, humidity, and wind speed through a grid-like layout. After targeted noise reduction and fusion by an image noise adaptive filtering module, the insect dual-branch feature analysis network simultaneously extracts insect morphology and spectral features, accurately identifying the species and quantity. Furthermore, it quantifies population density using an environmental correction coefficient, and then corrects and predicts short-term trends based on diffusion rate, classifying three levels of early warning. Finally, it generates precise control instructions and accurately assesses the control effect, significantly improving insect identification accuracy, predicting diffusion risks in advance, avoiding blind pesticide use and resource waste, adapting to different agricultural scenarios, significantly improving insect monitoring efficiency and control effectiveness, and contributing to ensuring safe crop production.

[0050] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, enabling those skilled in the art to better understand and utilize it. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A regional pest early warning and control system based on visual recognition, characterized in that, It includes a pest multispectral trapping imaging module, an image noise adaptive filtering module, a pest dual-branch feature analysis module, a pest population density quantification module, a pest spread trend prediction module, and a precision control instruction adaptation module. The multispectral trapping and imaging module for pests simultaneously acquires images of pests in different spectral bands and environmental parameters of the monitoring area; An adaptive image noise filtering module removes image noise and fuses multispectral images; a dual-branch feature parsing module for pests constructs a dual-branch feature parsing network to accurately identify and distinguish pests. The pest population density quantification module is used to quantify the pest population density; the pest spread trend prediction module predicts the short-term spread trend of pests and outputs the pest spread warning level; the precision control instruction adaptation module generates targeted precision control instructions.

2. The regional pest early warning and control system based on visual recognition according to claim 1, characterized in that, The first branch of the dual-branch feature parsing network is the morphological feature parsing branch, which extracts the morphological features of pests through convolutional layers and pooling layers, and uses an attention mechanism to focus on key parts of pests. The second branch of the dual-branch feature parsing network is the spectral feature parsing branch, which extracts the reflectance differences of insect pests in different spectral bands in the fused image and generates spectral feature vectors.

3. The regional pest early warning and control system based on visual recognition according to claim 1, characterized in that, The specific process for quantifying pest population density is as follows: The system receives parameters transmitted from the pest dual-branch feature analysis module and the pest multispectral trapping imaging module, and then calculates the environmental correction coefficient using environmental parameters. After obtaining the environmental correction coefficient k, the pest population density is calculated using the population density quantification formula. After quantification, the pest population density ρ and various parameters calculated during the process are transmitted to the pest spread trend prediction module.

4. The regional pest early warning and control system based on visual recognition according to claim 3, characterized in that, The specific process by which the pest spread trend prediction module predicts the short-term spread trend of pests is as follows: Based on the type of pest, the preset pest diffusion rate baseline value ro is called. The ro values ​​are different for different pests. Then, the diffusion rate baseline value is corrected in combination with environmental parameters to obtain the actual diffusion rate r of the corresponding pest. After calculating the actual diffusion rate r of the pest, the pest population density ρt after time t is predicted, and the pest population density increment Δρ of the corresponding pest within the prediction period is calculated. After the prediction is completed, the module classifies the pest warning level according to the size of Δρ, and transmits the pest spread prediction information, including the pest warning level, ρt and Δρ, to the precision prevention and control instruction adaptation module.

5. The regional pest early warning and control system based on visual recognition according to claim 4, characterized in that, The strategy for classifying pest warning levels is as follows: if Δρ≤0.5, it is marked as a Level 1 warning; if 0.5<Δρ≤1.5, it is marked as a Level 2 warning; if 1.5<Δρ, it is marked as a Level 3 warning.

6. The regional pest early warning and control system based on visual recognition according to claim 4, characterized in that, The precision control instruction adaptation module determines the control method based on the pest type, and determines the control intensity and scope based on the pest spread warning level and predicted population density. At the same time, it adjusts the concentration, dosage and application time of the control agent in combination with the crop type in the monitoring area, and generates detailed control instructions.

7. The regional pest early warning and control system based on visual recognition according to claim 6, characterized in that, The specific methods for determining prevention and control methods are as follows: for small piercing-sucking pests, biological control should be given priority; for large chewing pests, physical control combined with chemical control should be used; for pests that spread rapidly and have high density, chemical control should be the main method and biological control should be used as a supplement. The specific methods for determining the intensity and scope of prevention and control are as follows: for Level 1 warnings, localized prevention and control are carried out only in areas where pests are concentrated; for Level 2 warnings, routine prevention and control are carried out in the entire monitoring area; and for Level 3 warnings, the scope of prevention and control is expanded and the frequency of prevention and control is increased.

8. The regional pest early warning and control system based on visual recognition according to claim 7, characterized in that, Both the pest spread trend prediction module and the precision control instruction adaptation module are communicatively connected to the pest control evaluation and optimization module. The pest control evaluation and optimization module evaluates the pest control effect and calculates the pest control efficiency ηc. If the pest control efficiency ηc ≥ 80%, the pest control is deemed qualified; if the pest control efficiency ηc < 80%, the pest control is deemed unqualified.

9. The regional pest early warning and control system based on visual recognition according to claim 8, characterized in that, If pest control is unsatisfactory, analyze the reasons, including incorrect pest identification, deviation in spread prediction, and unreasonable control instructions, and calibrate the relevant module parameters accordingly.

10. A method for regional pest early warning and control based on visual recognition, characterized in that, Includes the following steps: Step 1: Pest spectral trapping and transportation; Step 2: Adaptive purification of insect image clutter; Step 3: Deconstruction of two-dimensional pest features; Step 4: Quantification of pest population density; Step 5: Prediction of pest spread trend; Step 6: Generation of precision control instructions; Step 7: Feedback on pest control.

Images