Intelligent monitoring and prediction system 3.0 for pests in tropical orchards

Abstract

The present application relates to a kind of tropical orchard pest intelligent monitoring and prediction system.The system includes pest trapping module, pest information holographic collection system, multi-model fusion AI identification system, solar photovoltaic power supply system and man-machine interface.Through pest trapping module, pest is trapped, pest pattern is photographed and collected using pest information holographic collection module, pest population dynamics is predicted through multi-module fusion AI identification system, the device is powered by solar photovoltaic power supply system, and user can carry out voice query and real-time viewing through man-machine interface.The application can improve the precision and predictability of tropical orchard pest control, effectively reduce pesticide dependence, and reduce the threshold for use, with important ecological and economic value.

Description

Technical Field

[0001] This invention relates to the fields of intelligent monitoring technology for agricultural pests, Internet of Things technology, and artificial intelligence, specifically to an intelligent monitoring and prediction system for tropical orchard pests that integrates computer vision, efficient trapping mechanical structure design, multimodal artificial intelligence algorithms, and biological mathematical models. Background Technology

[0002] Tropical regions, due to their unique climate, are key areas for the development of distinctive and efficient agriculture in my country, with industries such as dragon fruit and pineapple achieving significant planting scale. However, the persistent high temperature and humidity also lead to frequent and rapid outbreaks of pests and diseases, posing serious challenges to the accuracy and predictability of traditional monitoring and control methods, creating a technological gap that urgently needs to be filled.

[0003] Current agricultural pest monitoring products generally suffer from the following technical deficiencies: 1. Insufficient trapping specificity: Traditional equipment relies on a single pheromone bait, resulting in a false attraction rate of over 35% for non-target insects, which disrupts the ecological balance; 2. Limited recognition accuracy: Recognition schemes based on basic image processing typically have an accuracy of less than 90%, making it difficult to distinguish between pest species with similar morphology (such as citrus psyllids and whiteflies). 3. Lack of predictive capabilities: Existing technologies lack population dynamic prediction models that incorporate environmental parameters (temperature, humidity, light intensity), leading to delayed prevention and control. 4. Poor environmental adaptability: Under high temperature and high humidity conditions, the stability of pheromone release is poor and the equipment failure rate increases significantly.

[0004] 5. High technical barriers to operation: Traditional smart devices are complex to operate, farmers are on average over 50 years old, have limited digital literacy, and existing products lack graphical interfaces and voice interaction functions, resulting in insufficient usage.

[0005] In particular, outbreaks of pests in tropical orchards are characterized by their sudden onset and rapid population turnover, which further amplifies the difficulty of prevention and control.

[0006] Therefore, there is an urgent need for an intelligent pest monitoring system that combines targeted trapping, high-precision identification, dynamic prediction, and environmental robustness. Summary of the Invention

[0007] This invention aims to solve the technical problems of poor trapping specificity, low identification accuracy, lack of prediction function and weak environmental adaptability of traditional pest monitoring equipment, and provides an intelligent monitoring and prediction system for tropical orchard pests, which can realize fully automatic targeted trapping of pests, high-precision identification and population dynamic prediction.

[0008] (a) High specificity in trapping This invention employs a triple trapping mechanism combining mechanical, chemical, and physical methods to enhance efficiency and specificity. It integrates multiple technologies, including a triangular roller trapping module, intelligent pheromone release, and insect-attracting light with a specific spectrum, to achieve targeted attraction of pests. Specifically, it utilizes a triangular roller trapping module for efficient physical capture, while simultaneously incorporating intelligent pheromone release based on biological characteristics and insect-attracting light with a specific spectrum, forming a complete trapping chain of "physical barrier + chemical attraction + spectral attraction." Each trapping mechanism works collaboratively, and real-time feedback and adjustment are achieved through closed-loop control technology, ensuring precise trapping throughout the entire process and significantly reducing the false attraction rate of non-target insects.

[0009] (ii) High recognition accuracy This invention employs a multi-model fusion AI recognition system, comprehensively utilizing multiple algorithms such as the YOLOv11 target detection model, a self-supervised learning model, and a biomathematical prediction model to achieve accurate pest identification, quantity statistics, and trend prediction. The YOLOv11 target detection model adopts the latest YOLOv11 architecture, including a Backbone network, a Neck network, and a Head network. Through deep learning technology, it achieves precise pest location and species identification with an accuracy rate of no less than 98%, and the time from image acquisition to recognition result output is only 0.05 seconds. The self-supervised learning model employs five innovative tasks: rotation prediction, contrastive learning, mask reconstruction, color restoration, and texture analysis. This reduces the required labeled data by 90% and achieves a recognition accuracy of 98.8%. The various AI models work collaboratively to form a complete intelligent chain of "identification-verification-prediction," providing real-time feedback to the user through a human-computer interaction interface, realizing fully intelligent operation of the system.

[0010] (III) Scientific and accurate prediction This invention employs a biomathematical prediction model based on entomological principles. Through scientific calculations using temperature adaptability, humidity adaptability, pheromone efficacy, and population dynamics models, it achieves 7-day pest population prediction with an accuracy rate exceeding 90%. Specifically, it combines environmental parameters such as temperature, humidity, and pheromone efficacy to drive the biomathematical prediction model, enabling scientific prediction of pest population dynamics and outbreak risks. The system automatically generates corresponding control recommendations, including optimal control time windows, recommended control methods, and estimated pesticide dosages, which are presented to users through a human-computer interaction interface in the form of visual charts and text reports, effectively ensuring the scientific validity and accuracy of the system's predictions.

[0011] (iv) Strong environmental adaptability This invention employs a solar photovoltaic power supply system, which, through precise calculations and experimental verification, can meet the system's long-term stable power supply requirements. Specifically, it utilizes high-efficiency photovoltaic panels with dual-axis tracking capabilities, coupled with an intelligent controller integrating the MPPT algorithm to maximize solar energy collection efficiency. Simultaneously, the system incorporates a large-capacity energy storage unit and adopts a fully sealed protective structure, ensuring uninterrupted power supply even during continuous rainy weather and in high-temperature and high-humidity environments. The system casing features an IP67-rated waterproof and dustproof design, and all electrical interfaces are sealed, ensuring stable operation in tropical high-temperature and high-humidity environments and significantly improving the system's environmental adaptability.

[0012] (v) User-friendly human-computer interaction This invention fully considers the user's needs during use, integrating multi-terminal access and intelligent interaction functions to create an intuitive and convenient management portal. Specifically, it provides a visual operation panel across multiple platforms, including web pages and mobile applications, supporting remote real-time monitoring and device parameter settings. The interface also deeply integrates intelligent voice Q&A and early warning push mechanisms, transforming complex monitoring data and prediction results into understandable decision suggestions. All functional modules of the system achieve seamless integration with IoT services and a visual design. In addition to using the system's built-in controller, users can also remotely monitor via a computer webpage or a mobile app, significantly reducing the operational threshold and increasing usability.

[0013] This invention relates to an intelligent monitoring and prediction system for tropical orchard pests. The system includes a pest trapping module, a holographic pest information collection system, a multi-model fusion AI recognition system, a solar photovoltaic power supply system, and a human-computer interaction interface.

[0014] The pest trapping module includes a triangular roller insect trapping module, a pheromone releaser, an insect-attracting light strip, and a stepper motor.

[0015] The triangular roller insect-trapping module consists of a lower roller supplying the insect-trapping tape, a middle roller for trapping, and an upper roller for retrieving the tape. The three rollers are arranged in an equilateral triangle, secured at the top and bottom by left and right side support brackets, with the middle roller fixed to the sheet metal housing. The rollers are covered with sticky tape, and a stepper motor drives their rotation, allowing for periodic replacement of the sticky surface and ensuring consistent trapping effectiveness. The pheromone releaser uses a composite pheromone formula optimized for the sex and aggregation pheromones of common tropical orchard pests. Equipped with a high-precision liquid level sensor and an electrically controlled precision metering valve, it automatically adjusts the release dosage based on pest type and environmental conditions, and works in conjunction with a temperature control switch for intelligent and precise release. The insect-attracting light strip uses a specific wavelength LED light source, optimized for the spectral sensitivity of common tropical orchard pests. It features intelligent brightness adjustment, automatically adjusting the light intensity based on ambient light intensity. Working in conjunction with the pheromone releaser, it forms a dual trapping mechanism of "light attraction + chemical attraction," significantly improving pest trapping efficiency and specificity.

[0016] The holographic pest information collection system includes a microscope camera, a multi-functional motherboard, and an edge computing processing module. The microscope camera is equipped with a high-resolution CMOS image sensor with a pixel resolution of no less than 8 megapixels, featuring macro shooting and autofocus capabilities, enabling it to capture high-definition images of tiny pests. The multi-functional motherboard module integrates an ARM architecture processor, a 4G wireless communication module, a BeiDou positioning module, and a high-efficiency heat dissipation module. It is responsible for the initial processing, storage management, and system coordination control of image data, ensuring the stability and accurate positioning of the device during remote operation. The edge computing processing module is deployed on the Jetson Nano embedded platform and is used to perform preprocessing and feature extraction of pest images on the device side.

[0017] The multi-model fusion AI recognition system employs a YOLOv11 architecture, a self-supervised learning model, and a biomathematical prediction model. The YOLO recognition model uses the YOLOv11 architecture, with its core structure including an improved CSPNet (Cross Stage Partial Network) as the backbone for feature extraction, a PathAggregation Network for multi-scale feature fusion, and a decoupled head for final detection, enabling real-time identification of pest species and quantities on sticky tape. The self-supervised learning model employs four tasks: contrastive learning, mask reconstruction, color restoration, and texture analysis, significantly reducing the need for labeled data. The biomathematical prediction model, based on temperature adaptation equations, humidity adaptation equations, pheromone efficacy equations, and population dynamics equations, combined with historical pest data and environmental parameters, outputs the pest outbreak risk level for the next 7 days, providing a scientific basis for prevention and control decisions.

[0018] The solar photovoltaic power supply system includes solar photovoltaic panels, a solar voltage regulator, and a smart circuit breaker. The solar photovoltaic panels utilize high-efficiency monocrystalline silicon solar cells and are equipped with AI-driven intelligent light-tracking and temperature-tracking dual-mode functions to maximize solar energy utilization. The solar voltage regulator integrates fourth-generation MPPT (Maximum Power Point Tracking) technology to ensure efficient power conversion and stable output. The smart circuit breaker provides triple overload protection and instantaneous short-circuit protection to ensure system electrical safety and prevent equipment damage due to electrical faults.

[0019] The human-computer interaction interface includes a visual real-time monitoring interface and a remote control interface. The remote control interface supports mobile access, allowing users to adjust pheromone release strategies and view warning information via voice input, enabling convenient remote operation and intelligent management. The human-computer interaction interface adopts a responsive design, supporting access from multiple terminals including PCs, tablets, and mobile phones, ensuring a convenient user experience in any scenario.

[0020] The system operation method includes the following steps: Step 1: Pest Trapping The pheromone releaser automatically adjusts the pheromone release intensity based on ambient temperature and pest species, releasing specific pheromones for targeted trapping of different pests. The insect-attracting light strip automatically turns on or off according to ambient light intensity, providing optimal trapping effects during peak pest activity at night, forming a dual trapping mechanism of "light attraction + chemical attraction." Driven by a stepper motor, the triangular roller trapping module supplies fresh sticky insect tape to the lower roller, the middle roller serves as the trapping surface, and the upper roller retrieves the tape with insects attached, achieving a synergistic operation of the tape and pheromone trapping, with pests being adhered and captured by the sticky insect tape.

[0021] Step Two: Pest Information Collection A microscope camera is used to collect pest information data in the work area. In conjunction with a triangular roller insect-trapping module, the microscope camera scans and photographs the sticky insect tape at regular intervals or as needed. The collected pest information data includes pest morphological characteristics, quantity, location distribution, and capture time. Simultaneously, supporting environmental sensors collect environmental parameters such as temperature, humidity, and light intensity, as well as precise detection time, providing multi-dimensional data support for subsequent analysis.

[0022] Step 3: Preliminary Data Processing The multi-functional motherboard preprocesses the acquired image data, including image enhancement, noise filtering, and target segmentation. Image processing algorithms are used to initially identify the types and numbers of pests, generating a preliminary pest distribution map. Combining high-resolution images from the microscope camera with the coordinates of the sticky insect tape within the triangular roller system, the specific location of each pest is determined, forming a pest distribution dataset containing spatial information.

[0023] Step 4: AI Deep Recognition and Analysis The initially processed data is transmitted to a multi-model fusion AI recognition system. First, the YOLOv11 object detection model accurately locates and identifies pests in the image. Second, a self-supervised learning model validates and optimizes the recognition results, improving accuracy and robustness. Finally, deep analysis is performed based on the pest distribution map to obtain the final pest recognition results, including the quantity, distribution density, and spatial aggregation characteristics of various pests. Simultaneously, the specific locations of the pests are spatially mapped to generate a visualized pest distribution spatial map. The recognition results and distribution spatial map are then compared. Figure 1 And transmit it to the prediction model software system.

[0024] Step 5: Forecasting and Decision Support After receiving information on the location, quantity, and species of pests in various regions, the predictive model software system combines historical monitoring data with current environmental parameters and performs comprehensive analysis using a biomathematical prediction model. Based on temperature adaptability, humidity adaptability, pheromone efficacy, and population dynamics models, it calculates the pest population trend for the next 7 days and assesses the risk level. The system automatically generates corresponding control recommendations, including the optimal control time window, recommended control methods, and estimated pesticide dosage, which are presented to users through a human-computer interface in the form of visual charts and text reports to assist users in making pest management decisions.

[0025] The prediction model software system is equipped with a memory to store all data acquired by the microscope camera, including raw images, recognition results, pest distribution maps, environmental parameter data, and prediction results. This stored data serves a dual purpose: firstly, it is used to conduct a comprehensive and multi-faceted evaluation and analysis of the monitoring work after prediction, generating a monitoring quality report to help users understand the monitoring effectiveness; secondly, through built-in optimization prediction algorithms, the system can learn and extract patterns from historical data, continuously optimizing model parameters to improve the accuracy and reliability of future predictions, thus achieving system self-evolution.

[0026] Furthermore, considering the scientific rigor and accuracy of pest population prediction, the biomathematical prediction model comprises four interconnected sub-models: (1) Temperature adaptation model: T_adapt = exp(-((T-T_opt)²) / (2σ²)) Where T is the current ambient temperature (°C), T_opt is the optimal growth temperature (°C) for a specific pest, and σ is the temperature tolerance variance. This model describes the adaptability of a pest population to temperature changes. When the ambient temperature is close to the optimal temperature, the fitness coefficient is close to 1, and the pest's activity and reproductive capacity are strongest; the further away from the optimal temperature, the smaller the fitness coefficient.

[0027] (2) Humidity adaptability model: H_adapt = 1 / (1+exp(-k(H-H_threshold))) Where H represents the current ambient humidity (%), H_threshold is the humidity threshold (%), and k is the sensitivity coefficient. This model uses a sigmoid function to describe the response of pest populations to humidity changes. When the humidity is above the threshold, the fitness coefficient is high; when it is below the threshold, the fitness coefficient decreases rapidly.

[0028] (3) Pheromone efficacy model: P_effect = P_release × E_capture × C_diffusion Where P_release is the pheromone release intensity (mg / h), E_capture is the tape capture efficiency (%), and C_diffusion is the environmental diffusion coefficient. This model comprehensively evaluates the overall effectiveness of the pheromone trapping system, considering the three key stages of release, diffusion, and capture, and reflects the actual effect of pheromones on pest trapping.

[0029] (4) Population dynamics model: N(t) = K / (1+((K-N0) / N0)exp(-rt)) Where N(t) is the population size at time t (in insects), K is the environmental carrying capacity (in insects), N0 is the initial population size (in insects), and r is the intrinsic growth rate (1 / day). This model, based on the Logistic growth equation, describes the change in pest population size over time under limited resources. When the population size is much smaller than the environmental carrying capacity, it grows exponentially; as it approaches the environmental carrying capacity, the growth rate gradually slows down.

[0030] Comprehensive pest population prediction: N_predict(t) = N(t) × T_adapt × H_adapt × P_effect By combining the four sub-models, the system can scientifically predict the population size of pests at a specific time in the future based on the current pest population, environmental temperature and humidity conditions, and pheromone trapping effectiveness, providing a quantitative basis for precise prevention and control. Detailed Implementation

[0031] The present invention will now be described in detail with reference to specific embodiments.

[0032] Example 1: Application of pest monitoring in a dragon fruit orchard in Hainan The monitoring system of this invention was deployed at a dragon fruit plantation (approximately 50 acres) in Hainan Province. The system is installed in the center of the orchard and uses a triangular roller trapping system to continuously trap insects, covering a radius of 100 meters.

[0033] Preparations before system operation: The system is initialized via a remote control interface, and monitoring parameters are set: the shooting frequency is once every 2 hours, the pheromone release intensity is automatically adjusted according to the temperature (25-30℃ is the standard release intensity), and the sticky tape is replaced every 10 days. The intelligent human-machine interface supports Hainan dialect voice interaction, and the operation process is simplified to "one-click start". Video tutorials are provided to guide ordinary small farmers, enabling them to complete system deployment and daily operation without professional training.

[0034] System operation process: Phase 1 (Pheromone trapping and tape collection): After the system is started, the pheromone releaser automatically adjusts the release intensity according to the ambient temperature (25-35℃), releasing compound pheromones to attract pests. The Beidou positioning module records the system coordinates in real time (accuracy ±2 meters). The triangular roller system starts synchronously, with the lower roller providing fresh tape, the middle large roller serving as the pheromone trapping surface, and the upper roller collecting the tape with insects stuck to it, achieving continuous trapping operations.

[0035] Phase Two (Image Acquisition and Recognition): Every two hours, the system automatically triggers the monitoring process. A high-resolution optical imaging module (8-megapixel CMOS sensor), driven by a triangular roller drive module, scans and photographs the surface of the sticky insect tape on the central roller. The roller is driven by a stepper motor, rotating 30 degrees each time. The camera has an autofocus function, capable of clearly capturing tiny pests as small as 1-2 mm in length.

[0036] Image data is transmitted to the edge computing processing module for preprocessing (contrast enhancement, sharpening, background segmentation, etc.), pheromone release data and tape usage are recorded, and then uploaded to the cloud server via the 4G wireless communication module.

[0037] Phase Three (Forecasting and Decision-Making): The predictive model software system receives the identification results, retrieves environmental sensor data (current temperature 28℃, humidity 75%), pheromone release records, and triangular roller operation data, and calculates based on the biomathematical prediction model: 1) Temperature adaptability: Taking thrips as an example, T_opt=25℃, σ=3, T=28℃, the calculated T_adapt=0.8.

[0038] 2) Humidity adaptability: H_threshold=65%, k=0.2, H=75%, calculated to be H_adapt=0.88.

[0039] 3) Pheromone efficacy: P_effect = P_release × E_capture × C_diffusion = 0.85.

[0040] 4) Population dynamics: N0=127 individuals, K=2000 individuals, r=0.15, predicting N(7)=245 individuals after 7 days and N(14)=412 individuals after 14 days.

[0041] Overall prediction: N_predict(7) = 146 animals, N_predict(14) = 246 animals.

[0042] Phase Four (Intelligent Decision-Making and User Reminders): The system assesses the risk level (low, medium, high, and very high) based on the prediction results and generates prevention and control recommendations. When the thrips population is increasing but has not reached the prevention and control threshold (500 individuals), the risk level is "medium." The system sends an early warning to the user through an intelligent human-computer interaction interface, recommending the first prevention and control measure to be carried out in 7-10 days, with biological control methods as the priority.

[0043] Phase 5 (Data Display and Management): The intelligent human-machine interface updates monitoring and forecast results in real time. Users can view equipment parameters, pest identification results, forecast data, and environmental parameters through a mobile app. The system intelligently prompts that rainfall is expected in the next 3 days, suggesting preventative measures and checking tape collection.

[0044] System continuous operation effect: After three months of continuous monitoring, the pest control effect in the dragon fruit orchard was significant: the number of treatments was reduced from four times a month to two times, pesticide use was reduced by 60%, saving about 150 yuan per mu in pesticide costs; the fruit loss rate was reduced from 11% to 7%, reducing losses by about 300 jin per mu, and increasing income by about 1,000-1,500 yuan per mu; pheromone trapping efficiency was increased by 35%, the accidental trapping rate of neutral insects was reduced to below 5%, and damage from natural enemies was reduced by 90%; the system's investment payback period was shortened to 1.5 years.

[0045] Example 2: Grid-based intelligent scheduling of large-scale planting bases The intelligent monitoring system of this invention was deployed at a large mango plantation in Hainan Province (covering an area of ​​approximately 240 acres). Addressing the challenges of a large plantation with vast areas and complex terrain, the system employs a "grid-based deployment and collaborative operation" capability. It connects multiple monitoring points to a network, utilizing a distributed network architecture to achieve real-time data synchronization. Combined with cloud-based algorithms, the system performs data fusion analysis, generating a "pest distribution heat map" and a "risk level zoning map" for the entire plantation, thus upgrading from "single-point monitoring" to "regional joint prevention and control."

[0046] System configuration features: (1) Multi-device networking: The system is set up with 12 monitoring nodes. Each node is equipped with an independent high-resolution optical imaging module, edge computing processing module and triangular roller insect trapping module. Data is synchronized in real time through 4G+WiFi6 dual-mode communication technology, and the coverage rate reaches more than 90%.

[0047] (2) Intelligent path planning: The system automatically optimizes the deployment location of each monitoring node based on the orchard topography and historical data on pest distribution to ensure that the monitoring coverage reaches more than 90% and avoid monitoring blind spots.

[0048] (3) Distributed data fusion: The cloud server receives data from 12 monitoring nodes in real time and generates a three-dimensional heat map of pest distribution in the entire park through a spatiotemporal fusion algorithm, so as to achieve accurate spatial analysis.

[0049] System operation process: Phase 1 (Collaborative Monitoring and Data Acquisition): Each monitoring node synchronously initiates pheromone release and image acquisition, with a time error controlled within one second. Each node is configured with a specific pheromone formula for common mango orchard pests (thrips, aphids, whiteflies, fruit flies, etc.), achieving nano-level precision release through microfluidic chip technology. The triangular roller insect-trapping module employs three precision roller mechanisms arranged in an equilateral triangle, enabling coordinated operation of light strips, tape, and pheromone trapping. The surface of the large central roller is coated with the color spectrum and texture most sensitive to the pest, forming a dual "odor + visual" bait.

[0050] Phase Two (Edge Computing and Cloud Integration): Each monitoring node's edge computing processing module performs real-time preprocessing on the collected image data (image enhancement, noise filtering, target segmentation, etc.), uses image processing algorithms to preliminarily identify the types and quantities of pests, generates a preliminary pest distribution map, and then uploads the processed images to the cloud server via a 4G wireless communication module. After receiving data from 12 monitoring nodes, the cloud server generates a 3D heat map of pest distribution across the entire park using a spatiotemporal fusion algorithm, enabling precise spatial analysis. The system performs horizontal comparison and vertical analysis of pest data from the 12 monitoring points, generating a spatial distribution difference report and trend prediction map.

[0051] Phase Three (Intelligent Analysis and Regional Joint Defense): Based on 3D heat maps and spatial distribution difference reports, the system automatically identifies the aggregation areas and spread patterns of pests. Monitoring revealed that the number of pests at monitoring points near the edge of the forest was significantly higher than that in the center of the park, with a difference coefficient of 3.2 times, providing a scientific basis for precise prevention and control.

[0052] The system uses a biomathematical prediction model to perform a comprehensive analysis, calculate the pest population trend over the next 7-100 days, and assess the risk level. The system automatically zons the risk level, dividing the entire park into low-risk, medium-risk, high-risk, and extremely high-risk zones, and generates corresponding prevention and control recommendations.

[0053] Phase Four (Collaborative Decision-Making and Unified Scheduling): Based on the risk level zoning map, the system sends regional joint prevention and control recommendations to users. For high-risk areas, the system recommends immediate prevention and control measures; for medium-risk areas, the system recommends taking prevention and control measures after 7-10 days; for low-risk areas, the system recommends continued monitoring.

[0054] The system also supports unified scheduling, allowing users to adjust the pheromone release strategy, shooting frequency, and conveyor belt speed of all monitoring nodes through an intelligent human-machine interface, thus achieving regional joint prevention and control.

[0055] System performance: (1) It has achieved full coverage intelligent monitoring of 240 mu of mango orchard, greatly improving the monitoring accuracy and efficiency, increasing the accuracy rate by 85% compared with manual monitoring, and reducing the monitoring blind area by 89%.

[0056] (2) Through multi-node data fusion and spatial distribution analysis, the aggregation areas and spread patterns of pests were successfully discovered, and the prediction accuracy reached 90%, providing strong scientific support for regional joint prevention and control.

[0057] (3) The system has been running continuously and stably for 6 months, with a failure rate of less than 5% and maintenance costs of only 20% of traditional manual monitoring, resulting in significant economic benefits.

[0058] (4) Through regional joint prevention and control, the prevention and control efficiency has been increased by 40% and the amount of pesticides used has been reduced by 50%, realizing the upgrade from "single point monitoring" to "regional joint prevention and control".

[0059] Example 3: Specialized Optimization of Specialty Crops For different tropical crops such as mango, lychee, and dragon fruit, whose core pests and disease patterns are different, this invention achieves precise prevention and control of "specialized projects for specific fruits" by changing the pheromone formula, fine-tuning the AI ​​recognition model, and calibrating the parameters of the biomathematical prediction model.

[0060] System configuration features: (1) Crop-specific pheromone formula library: The system has a built-in pheromone formula library for 10 tropical crops such as mango, lychee, and dragon fruit. Each crop is configured with a specific pheromone formula for its core pest (such as mango mainly targeting thrips and lychee mainly targeting stink bugs), and the pheromone formula is precisely released through microfluidic chip technology.

[0061] (2) Crop-specific AI recognition model: The system has specially trained a dedicated YOLO11 target detection model for the core pests of each tropical crop, and increased the training data for the crop pests, which improves the recognition accuracy by 2-3%.

[0062] (3) Crop-specific biomathematical prediction model: The system has specifically calibrated the parameters of the biomathematical prediction model for the core pests of each tropical crop, such as adjusting the optimal temperature T_opt, humidity threshold H_threshold, intrinsic growth rate r, etc., which improves the prediction accuracy by 5-8%.

[0063] System operation process: Phase 1 (Crop Identification and Formula Switching): After the system starts up, it first identifies the type of crop currently planted (e.g., mango) through user input or automatic identification. Then, it retrieves the mango-specific pheromone formula from the crop-specific pheromone formula library and uses microfluidic chip technology to achieve precise nano-level release. The system simultaneously switches to a mango-specific AI recognition model, which has been specially trained for mango's core pests (thrips, aphids, whiteflies, fruit flies, etc.), improving recognition accuracy by 2-3%.

[0064] Phase Two (Image Acquisition and Specific Recognition): The system automatically triggers a monitoring process every 2 hours. Driven by the triangular roller transmission module, the high-resolution optical imaging module scans and captures images of the sticky insect tape on the central roller. After capturing the images, the data is transmitted to the edge computing processing module, where Mango's proprietary AI recognition model accurately locates and identifies the pests in the images, improving recognition accuracy by 2-3%.

[0065] Phase Three (Personalized Forecasting and Precise Decision-Making): The system uses a mango-specific biomathematical prediction model for comprehensive analysis to calculate the pest population trends over the next 7-100 days. The mango-specific biomathematical prediction model has parameters specifically calibrated for core mango pests (such as thrips), including optimal temperature T_opt = 27℃ (literature value 25℃, an 8% improvement), humidity threshold H_threshold = 70% (literature value 65%, an 8% improvement), intrinsic growth rate r = 0.18 (literature value 0.15, a 20% improvement), resulting in a 5-8% improvement in prediction accuracy.

[0066] The system automatically assesses the risk level based on the prediction results and generates corresponding prevention and control recommendations. For mango thrips, the system recommends the first control measure to be carried out in 7-10 days, prioritizing biological control methods. It also suggests checking the pheromone releaser's working status to ensure effective trapping.

[0067] System performance: (1) Through precise control of "specialized fruit and special project", the accuracy of mango thrips identification has been increased to 90%, the accuracy of prediction has been increased to 90%, and the control effect has been significantly improved.

[0068] (2) The efficiency of pheromone trapping is increased by 40%, the capture rate of target pests is increased by 3-5 times, the accidental capture rate of neutral insects is reduced to less than 5%, the damage from natural enemies is reduced by 90%, and the ecological safety is greatly improved.

[0069] (3) The amount of pesticides used is reduced by 50%, the fruit loss rate caused by pests is reduced from 10% to 5%, the yield per mu is increased by about 200 jin, and the income per mu is increased by about 1,500 yuan according to market price.

[0070] Example 4: Perception-Decision-Execution Closed Loop This invention innovatively integrates hardware innovations (such as the triangular roller insect-catching module and microfluidic pheromone release) with software algorithm innovations (such as the YOLO11 target detection model, multimodal self-supervised learning model, and biomathematical prediction model) to form a closed-loop control system of "perception-decision-execution" and achieve a synergistic effect of 1+1>2.

[0071] Detailed description of the closed-loop process: Phase 1 (Perception): The YOLO11 target detection model identifies a surge in the number of pests. The system scans and captures images of the sticky insect tape using a high-resolution optical imaging module, transmitting the image data to an edge computing processing module. The YOLO11 target detection model accurately locates and identifies pests in the images, achieving an accuracy rate exceeding 90%. When the system detects a surge in the number of a certain pest (e.g., thrips increasing from 127 to 245, a 90% increase), it immediately triggers an early warning mechanism.

[0072] Phase Two (Decision-Making): Biomathematical prediction models determine that the outbreak threshold will be reached in 3 days. Based on the identification results of the YOLO11 target detection model, the system retrieves environmental sensor data (current temperature 28℃, humidity 75%), pheromone release records, and triangular roller operation data. The biomathematical prediction model performs a comprehensive analysis based on four interrelated sub-models: temperature adaptability model, humidity adaptability model, pheromone efficacy model, and population dynamics model, calculating the pest population trend for the next 3-7 days. Taking thrips as an example, the system predicts that the thrips population will reach 350 in 3 days, exceeding the outbreak threshold (300), classifying the risk level as "high," and immediately generates control recommendations.

[0073] Phase Three (Execution): The system automatically instructs the multi-channel pheromone intelligent release module to increase the release intensity. Based on the decision-making results of the biomathematical prediction model, the system immediately sends instructions to the multi-channel pheromone intelligent release module to automatically increase the pheromone release intensity for the specific pest, thereby enhancing the trapping effect. Simultaneously, the system sends precise early warning information to the user through an intelligent human-computer interaction interface, including pest species, current numbers, predicted trends, risk levels, and control recommendations.

[0074] Phase 4 (Feedback): The system continuously monitors the trapping effect and optimizes the strategy. The system continuously monitors the trapping effect of the multi-channel pheromone intelligent release module, acquiring pest capture data in real time through a high-resolution optical imaging module. When the system detects a significant improvement in trapping effect (e.g., thrips capture rate increasing from 30% to 60%), it automatically records the effectiveness of the strategy and uses it to optimize the parameters of the biomathematical prediction model. When the system detects an insignificant trapping effect, it automatically adjusts the pheromone release strategy, such as changing the pheromone formula or adjusting the release intensity, to ensure that the trapping effect is always at its optimal level.

[0075] Closed-loop synergistic effect: Through a closed-loop control system of "perception-decision-execution-feedback," the system achieves a deep integration of hardware innovation and software algorithm innovation, forming a synergistic effect of 1+1>2: hardware innovation (triangular roller insect trapping module, microfluidic pheromone release) provides the foundation for precise trapping; software algorithm innovation (YOLO11 target detection model, multimodal self-supervised learning model, biomathematical prediction model) provides the core of intelligent decision-making; the closed-loop control system deeply integrates hardware innovation and software algorithm innovation, realizing automated and precise prevention and control, improving trapping efficiency by 3-5 times and prediction accuracy by 5-8%.

[0076] Example 5: Data Flow and Model Iteration This invention innovatively utilizes data generated during daily operation (pest images, environmental data, and deviations between prediction results and actual conditions) for self-supervised learning, thereby enabling the AI ​​model to continuously optimize in real-world use and achieve a self-evolutionary capability that "gets smarter with use," demonstrating the long-term value of the system.

[0077] Detailed description of the data flow: Phase 1 (Data Acquisition): The system continuously generates multimodal data. During daily operation, the system continuously generates multimodal data, including: pest image data (the high-resolution optical imaging module acquires pest images every 2 hours, generating 12 images per day and 360 images per month), environmental data (environmental sensors collect data such as temperature, humidity, and light intensity in real time, once every 30 seconds, generating approximately 2,880 sets of data per day and approximately 86,400 sets of data per month), pheromone release data (the multi-channel intelligent pheromone release module records the pheromone release time, type, and dosage data in real time, generating 1 set of data for each release, generating approximately 4,000 sets of data per month), and prediction result data (the biomathematical prediction model generates 1 set of prediction results per day, including the predicted trend of pest numbers for the next 7-100 days, generating 30 sets of prediction results per month).

[0078] Phase Two (Data Labeling and Bias Analysis): The system automatically performs data labeling. The system utilizes multimodal self-supervised learning technology to automatically label the collected pest image data, reducing the required labeling data by 90%. The system also compares and analyzes the predicted results with the actual situation, calculating the prediction bias. For example, the system predicted 146 thrips after 7 days, while the actual monitoring result was 150, resulting in a prediction bias of 3%.

[0079] Phase 3 (Model Iteration and Optimization): The system automatically optimizes the model. Based on prediction deviation data, the system automatically optimizes the parameters of the biomathematical prediction model. For example, when the prediction deviation remains high, the system automatically adjusts parameters such as the optimal temperature T_opt, humidity threshold H_threshold, and intrinsic growth rate r to improve prediction accuracy. Simultaneously, the system uses newly acquired pest image data to incrementally train the YOLO11 target detection model, enabling the model to continuously learn new pest features and continuously improve recognition accuracy.

[0080] Phase Four (Self-Evolution): The system becomes "smarter the more it's used" After three months of continuous operation, the system accumulated approximately 1,080 pest images, 259,200 sets of environmental data, 12,960 sets of pheromone release data, and 90 sets of prediction results. Using this data, the system underwent three iterations of model optimization. The recognition accuracy of the YOLO11 target detection model improved from 85% to 90%, and the prediction accuracy of the biomathematical prediction model improved from 87% to 90%, demonstrating its self-evolving capability of becoming "smarter with use."

[0081] Example 6: Reliability Verification The system of this invention has undergone rigorous reliability verification tests, including continuous operation tests under extreme weather conditions, to ensure the stability and reliability of the system in practical applications.

[0082] Test location: A mango plantation in Hainan Province (approximately 100 mu). Testing period: 90 consecutive days (including rainy and dry seasons) Test equipment: 5 monitoring systems, deployed in a grid pattern. Testing process: Phase 1 (First 30 days): Normal weather conditions Weather conditions: Temperature range 25-32℃, humidity range 65-75%, mostly sunny. System operation status: The solar photovoltaic panels charge for 8-10 hours daily, maintaining battery power above 80% and ensuring stable power supply; the high-resolution optical imaging module acquires 12 pest images daily, with clear image quality and a 90% recognition accuracy; the triangular roller insect trapping module operates 24 hours daily, with stable tape transmission and good trapping effect; the multi-channel pheromone intelligent release module automatically adjusts the release intensity according to the ambient temperature, ensuring precise release and improving trapping efficiency by 35%; the edge computing processing module operates in online mode, with normal data transmission and no disconnections.

[0083] Phase Two (Days 31-60): Extreme Weather Conditions Weather conditions: 15 consecutive days of heavy rain, temperature range 22-28℃, humidity range 85-95%. System operation status: The daily charging time of the solar photovoltaic panels is reduced to 2-3 hours, and when the battery level drops to 50-60%, the system automatically activates the energy-saving mode, extending the battery life by 40%; the high-resolution optical imaging module collects 12 pest images daily, and although the image quality has slightly decreased, the recognition accuracy remains above 98%; the triangular roller insect trapping module operates 24 hours a day, with stable tape transmission and good trapping effect; the multi-channel pheromone intelligent release module automatically adjusts the release intensity according to the ambient temperature, ensuring accurate release and stable trapping efficiency; the edge computing processing module operates in online mode, with normal data transmission and no disconnection issues.

[0084] Phase 3 (Days 61-90): Weather returns to normal During this period, the weather gradually returned to normal, with temperatures ranging from 25-32℃ and humidity from 65-75%. System operation status: the daily charging time of the solar photovoltaic panels returned to 8 hours, battery power returned to over 80%, and power supply was stable; the high-resolution optical imaging module collected 12 pest images daily, with image quality returning to clarity and recognition accuracy returning to 90%; the triangular roller insect trapping module operated 24 hours a day, tape transmission returned to stability, and the trapping effect returned to good; the multi-channel pheromone intelligent release module automatically adjusted the release intensity according to the ambient temperature, release accuracy returned, and trapping efficiency returned to 35% improvement; the edge computing processing module returned to online working mode, data transmission returned to normal, and there were no disconnections.

[0085] Test results: After 90 days of continuous trouble-free operation testing, the system performed excellently: the equipment failure rate was 0%, with no hardware failures or software crashes; the system could still operate stably under extreme weather conditions, with stable power supply, no data loss, and normal core functions; the system automatically activated energy-saving mode, extending the battery life by 40%, ensuring stable operation even under extreme weather conditions; the system's average recognition accuracy was 90%, and the prediction accuracy was 89%, fully meeting the needs of practical applications; the system operated continuously for 90 days with zero maintenance costs, requiring only normal pheromone replenishment and tape replacement, with maintenance costs only 20% of those of traditional manual monitoring.

[0086] Example 7: Maintainability Implementation The system of this invention adopts a modular intelligent design, and each functional module implements a standardized interface design. Key components support hot-swappable replacement, which greatly reduces maintenance difficulty and cost.

[0087] Pheromones replenishment process: The system uses multi-sensor fusion technology to monitor pheromone release intensity and consumption rate in real time. When the pheromone concentration falls below the optimal threshold of 15%, the intelligent human-machine interface intelligently pushes a replenishment reminder and automatically calculates the optimal replenishment time. Users can replenish highly concentrated pheromones through the quick-fill port of the pheromone releaser. The operation steps are as follows: Open the protective cap of the pheromone releaser to expose the quick-fill port; use a dedicated applicator to inject the highly concentrated pheromone into the quick-fill port, according to the system prompts (e.g., 50 ml); close the protective cap, and the system automatically detects the pheromone concentration to confirm successful replenishment. The operation takes approximately 3 minutes, which is 40% more efficient than traditional methods. No professional technicians are required, and ordinary farmers can complete the task.

[0088] Tape replacement procedure: The system dynamically optimizes the replacement cycle (typically 8-16 days) based on intelligent algorithms considering pest capture volume, environmental conditions, and tape stickiness. When the system detects that the tape needs replacing, the intelligent human-machine interface intelligently pushes a replacement reminder. Users simply open the system casing, manually remove the used triangular roller insect-trapping module tape, and install a new tape reel. The steps are as follows: Open the maintenance door of the system casing to expose the triangular roller insect-trapping module; loosen the fixing screws of the tape reel and remove the used tape reel; install a new tape reel, adjust the tape tension to ensure stable tape transmission; close the maintenance door, and the system automatically detects the tape installation status and confirms successful replacement. The operation takes approximately 6 minutes, improving efficiency by 33% compared to traditional sticky insect boards, and requires no special tools, making it easy for ordinary farmers to complete.

[0089] Intelligent cleaning system: The system casing features an IP67-rated waterproof and dustproof design and is equipped with an auxiliary cleaning function. A built-in mini air compressor periodically cleans the high-resolution optical imaging module lens and the surface of the solar photovoltaic panel. It also intelligently detects the level of contamination and reminds the user to perform deep cleaning, ensuring optimal shooting quality and charging efficiency. The cleaning process is as follows: The system automatically detects the level of lens contamination; when the contamination level exceeds a threshold, a cleaning reminder is pushed through the intelligent human-machine interface. The user activates the automatic cleaning function, and the mini air compressor cleans the lens in approximately 30 seconds. The system automatically detects the level of contamination on the solar photovoltaic panel surface; when the contamination level exceeds a threshold, a cleaning reminder is pushed through the intelligent human-machine interface. The user cleans the solar photovoltaic panel surface with a soft cloth and water in approximately 5 minutes.

[0090] 1. It exhibits strong specificity in trapping and good ecological compatibility. This invention utilizes microfluidic chip technology to achieve intelligent multi-channel pheromone release, releasing specific pheromones for different pests. Combined with insect-attracting LED strips using specific wavelengths of LED light, it forms a dual trapping mechanism of "light attraction + chemical attraction." The pheromone releaser automatically adjusts the release intensity based on ambient temperature and pest species, while the insect-attracting LED strips automatically turn on or off according to ambient light intensity, providing optimal trapping effects during peak pest activity at night. The three-layer synergistic structure of the triangular roller trapping module enables coordinated operation of the tape and pheromone trapping, significantly reducing the false attraction rate of non-target insects, completely solving the problem of large-scale accidental injury to non-target insects caused by traditional equipment, and improving ecological compatibility.

[0091] 2. High recognition accuracy, adaptable to complex environments. This invention integrates the YOLOv11 deep learning model with self-supervised learning to improve pest identification accuracy in complex environments. YOLOv11 employs a three-level Backbone-Neck-Head network structure to achieve sub-pixel-level localization of pests in complex orchard environments. The self-supervised learning model, through five tasks—rotation prediction, contrastive learning, mask reconstruction, color restoration, and texture analysis—reduces the need for labeled data by 90% and achieves a recognition accuracy of 95%, overcoming the bottleneck in distinguishing morphologically similar insect species (such as the citrus psyllid and the whitefly) and establishing a dedicated identification knowledge system for tropical pests.

[0092] 3. Accurate and scientific forecasting enables early warning. This invention combines a biomathematical prediction model based on environmental parameters to achieve early warning of pest population dynamics. The biomathematical prediction model comprises four sub-models: a temperature adaptation model, a humidity adaptation model, a pheromone efficacy model, and a population dynamics model. The temperature adaptation model quantifies pest activity hotspots, the humidity response model reveals critical reproductive thresholds, the pheromone efficacy model optimizes trapping parameters, and the population dynamics model extrapolates growth trajectories. By collaboratively outputting the pest population change trend for the next 7 days through multi-model synergy, and conducting risk level assessments, this invention achieves a paradigm shift from passive extermination to proactive defense in pest control.

[0093] 4. Strong environmental adaptability and stable and reliable operation. This invention employs solar power supply and a sealed design to enhance the system's robustness in high-temperature and high-humidity environments. The solar photovoltaic power supply system uses high-efficiency monocrystalline silicon solar panels with a conversion efficiency of no less than 20%, equipped with AI-driven intelligent light-tracking and temperature-tracking dual-mode functions. The voltage regulator integrates MPPT maximum power point tracking technology, and the intelligent circuit breaker provides overload and short-circuit protection. The system casing features an IP67-rated waterproof and dustproof design, all electrical interfaces are sealed, and a wide-temperature-range heat dissipation design ensures uninterrupted system operation under extreme conditions such as tropical high-temperature and high-humidity environments and continuous rainy seasons.

[0094] 5. Low learning curve and user-friendly interface. This invention lowers the operational threshold and effectively improves usage through graphical interfaces and voice interaction. The human-computer interaction interface is designed for four platforms: web, mobile, WeChat mini-program, and intelligent voice interaction, employing a responsive design and supporting access from multiple terminals including PCs, tablets, and mobile phones. The real-time monitoring interface uses data visualization design, including dynamic pest species distribution maps, real-time quantity statistics charts, environmental parameter monitoring panels, and equipment operating status display modules. Users can adjust pheromone release strategies and view early warning information via voice input on the remote control interface, ensuring a user-friendly operating experience for smallholder farmers.

[0095] 6. Significant economic benefits and high promotional value. This invention significantly reduces equipment investment and maintenance costs through technological innovation, thereby improving overall economic benefits. Compared to traditional manual monitoring and experience-based control, this invention significantly improves the accuracy of pest monitoring and the scientific rigor of prediction, drastically reducing control costs and environmental pollution, and avoiding the problems of overuse and underuse of pesticides. Pesticide usage is reduced by 50-60%, labor costs are reduced by 70%, pest-induced losses are reduced from 10-15% to 3-5%, the rate of high-quality fruit increases by 15-20%, and income per mu (unit of land area) increases by 500-1000 yuan. The widespread application of this invention can not only improve the yield and quality of agricultural products and increase farmers' income, but also protect the ecological environment and promote sustainable agricultural development.

[0096] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention. Attached Figure Description

[0097] Figure 1 This is a schematic diagram of the overall three-dimensional structure of the intelligent monitoring and prediction system for tropical orchard pests in this invention. Figure 2 This is a front view of the overall intelligent monitoring and prediction system for tropical orchard pests in this invention. Figure 2 The middle one is a rainproof solar panel. Figure 2 The middle 2 is a 57 series two-phase hybrid stepper motor. Figure 2 The middle 3 is the shaft support frame. Figure 2 The middle 4 is a sheet metal shell. Figure 2 The fifth one is a microscope camera. Figure 2 The middle 6 is an insect trap. Figure 2 The seventh one is an insect trap. Figure 2 The middle 8 is the battery module. Figure 3 for Figure 1 Schematic diagram of the three-dimensional structure of the microscopic detection and imaging system in China Figure 4 for Figure 1 Schematic diagram of the three-dimensional structure of the pest trapping module Figure 5 for Figure 1 Engineering diagram of the pest trapping module Figure 6 for Figure 1 Simplified diagram of a solar photovoltaic power supply system Figure 7 for Figure 1 Workflow of Multi-Model Fusion AI Recognition System Figure 8 for Figure 1 Human-Computer Interaction Interface (HCI) Work Content Figure 9 for Figure 1 Working principle of human-computer interaction interface Appendix 1: System Core Component List 1. Microscope CMOS sensor 2. Motherboard Module ARM Processor 3. Triangular roller stepper motor drive 4. Monocrystalline silicon material for solar panels 5. Insect-attracting lamps and pheromone storage containers 6. High-sensitivity temperature control switch for pheromone releaser 7 System Enclosure Protection Design

[0098] To enable those skilled in the art to fully understand the technical solutions of the present invention, the present invention will be further described below in conjunction with embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.

[0099] The intelligent monitoring and prediction system for tropical orchard pests in this embodiment includes a pest trapping module, a pest information holographic collection system, a multi-model fusion AI recognition system, a solar photovoltaic power supply system, a prediction model software system, and a human-computer interaction interface. The insect trapping module includes a triangular roller insect trapping module, a pheromone releaser, an insect-attracting light strip, and a stepper motor. The lower roller of the triangular roller insect trapping module supplies the insect-trapping tape, the middle roller performs the trapping action, and the upper roller retrieves the tape. The three rollers are arranged in an equilateral triangle, fixed at the top and bottom by support plates on the left and right sides, with the middle roller fixed to the sheet metal shell. The pheromone releaser uses a composite pheromone formula, optimized for the sex pheromones and aggregation pheromones of common tropical orchard pests. It is equipped with a high-precision liquid level sensor and an electrically controlled precision metering valve, automatically adjusting the release dosage according to the pest species and environmental conditions, and achieving intelligent and precise release with a temperature control switch. The insect-attracting light strip uses a specific wavelength LED light source, optimized for the spectral sensitivity of common tropical orchard pests, and has an intelligent brightness adjustment function, automatically adjusting the light brightness according to the ambient light intensity. It works in conjunction with the pheromone releaser to form a dual trapping mechanism of "light attraction + chemical attraction." The stepper motor drives the rollers to rotate, enabling the periodic replacement of the insect-trapping surface.

[0100] The pest information holographic collection system includes a high-resolution optical imaging module, an edge computing processing module, and a multi-channel pheromone intelligent release module. The high-resolution optical imaging module is equipped with a high-resolution CMOS image sensor with a pixel resolution of no less than 8 million pixels. It features macro shooting, autofocus, and optical image stabilization to ensure clear capture of pest details. It integrates multi-sensor data fusion technology, using a Kalman filter algorithm to achieve real-time fusion processing of multi-sensor data. The edge computing processing module integrates an ARM architecture processor, a 4G wireless communication module, a BeiDou positioning module, and an intelligent temperature control and heat dissipation module. It is responsible for real-time processing, storage management, and system coordination control of image data, achieving millisecond-level real-time monitoring through 4G+WiFi6 dual-mode communication technology. The multi-channel pheromone intelligent release module uses microfluidic chip technology to achieve precise regulation and targeted release of pheromones from different pest species. It supports specific pheromone formulations for 15 common tropical orchard pests and is equipped with a high-sensitivity temperature control switch to automatically control the pheromone release intensity according to changes in ambient temperature.

[0101] The multi-model fusion AI recognition system adopts a multi-layered architecture, including a YOLOv11 target detection model, a self-supervised learning model, an intelligent dialogue system, and a biomathematical prediction model. The YOLOv11 target detection model uses the latest YOLOv11 architecture, including a Backbone network, a Neck network, and a Head network. Through deep learning technology, it achieves precise pest location and species identification with an accuracy rate of no less than 98%, and the time from image acquisition to output of the recognition result is only 0.05 seconds. The self-supervised learning model employs five innovative tasks: rotation prediction, contrastive learning, mask reconstruction, color restoration, and texture analysis, significantly reducing the need for labeled data and significantly improving recognition accuracy. The intelligent dialogue system supports natural language interaction, allowing users to communicate with the system via voice or text and obtain professional pest control advice. The biomathematical prediction model establishes a multi-dimensional pest population prediction model based on entomological principles, including a temperature adaptation model, a humidity adaptation model, a pheromone efficacy model, and a population dynamics model. These four models work synergistically to support medium- and long-term predictions with an accuracy rate of no less than 90%.

[0102] The solar photovoltaic power supply system includes solar photovoltaic panels, a solar voltage regulator, and a smart circuit breaker. The solar photovoltaic panels utilize high-efficiency monocrystalline silicon solar cells with a conversion efficiency of no less than 20%, and are equipped with automatic sun tracking to maximize solar energy utilization. The solar voltage regulator integrates MPPT (Maximum Power Point Tracking) technology, using intelligent algorithms to track the maximum power point of the solar panels in real time, ensuring efficient energy conversion and stable output. It also features overvoltage protection, short-circuit protection, and lightning protection. The smart circuit breaker provides overload and short-circuit protection to ensure system electrical safety and prevent equipment damage caused by electrical faults.

[0103] The predictive model software system includes a front-end user interface (UI) and a back-end model and database. The front-end UI is developed based on Web technology and supports access from multiple terminals. The back-end model and database integrate time series analysis and machine learning algorithms to establish a predictive model for pest population changes. It is equipped with a high-performance database system to store historical monitoring data and analysis results, and adopts a hybrid architecture of MySQL and MongoDB, with MySQL storing structured data and MongoDB storing unstructured data.

[0104] The human-machine interface includes a real-time monitoring interface and a remote control interface. The real-time monitoring interface adopts a visual data display design, including a dynamic pest species distribution map, real-time quantity statistics charts, an environmental parameter monitoring panel, and a device operation status display module, supporting simultaneous monitoring of multiple devices. The remote control interface is used for emergency shutdown and parameter adjustment of the system in case of emergencies, including functions such as pheromone release control, shooting frequency adjustment, and tape running speed adjustment.

[0105] The intelligent monitoring and prediction system for tropical orchard pests includes the following steps: Pests in the work area are attracted and captured using the system's pheromone releaser and insect-attracting light strip. The pheromone releaser automatically adjusts the pheromone release intensity according to the ambient temperature and pest species. The insect-attracting light strip automatically turns on or off according to the ambient light intensity. A triangular roller insect-trapping module, driven by a stepper motor, achieves coordinated operation of the adhesive tape and pheromone trapping, with pests being adhered and captured by the sticky tape. The system's high-resolution optical imaging module collects pest information data in the work area, including pest species, quantity, location distribution, environmental parameters, pheromone response data, and detection time. An edge computing processing module processes the pest species and quantity to obtain a pest distribution map. Finally, combining the image information obtained from the high-resolution optical imaging module, a pest distribution map is obtained. The system collects information on the distribution and specific locations of pests. Combining a multi-model fusion AI algorithm, it processes the pest distribution map to obtain an ideal pest identification result, and processes the specific locations of pests to obtain an ideal pest distribution map. Both are then transmitted to the prediction model software system. The prediction model software system obtains the location, quantity, and species of pests in each area, determines a reasonable prediction result, generates corresponding prevention and control suggestions, and transmits them to the human-computer interaction interface to guide pest management decisions. The prediction model software system is equipped with a memory to store the data acquired by the high-resolution optical imaging module, primarily the pest distribution map model established based on the pest information data. This allows for comprehensive and multi-faceted evaluation of the monitoring work after prediction, and, if the system has a dedicated optimization prediction algorithm, it can be used to optimize the next prediction.

[0106] Furthermore, considering the prediction of pest populations, the biomathematical prediction model includes: Temperature adaptation model: T_adapt = exp(-((T-T_opt)²) / (2σ²)), where T is the ambient temperature, T_opt is the optimum temperature, and σ is the temperature tolerance variance; Humidity adaptation model: H_adapt = 1 / (1+exp(-k(H-H_threshold))), where H is the ambient humidity, H_threshold is the humidity threshold, and k is the sensitivity coefficient; Pheromone efficacy model: P_effect = P_release × E_capture × C_diffusion, where P_release is the pheromone release intensity (mg / h), E_capture is the tape capture efficiency (%), and C_diffusion is the environmental diffusion coefficient; Population dynamics model: N(t) = K / (1+((K-N0) / N0)exp(-rt)), where N(t) is the population size at time t, K is the environmental capacity, N0 is the initial population size, and r is the intrinsic growth rate.

[0107] See Figure 4 and Figure 5 The most important component of the pest trapping module, namely the triangular roller insect trapping module, includes three precision roller mechanisms arranged in an equilateral triangle. The lower roller supplies the insect-trapping tape, the middle roller performs the trapping action, and the upper roller retrieves the tape. The three rollers are fixed vertically and horizontally by support plates on both sides, with the middle roller fixed to the sheet metal shell. The pheromone releaser uses a composite pheromone formula, optimized for the sex pheromones and aggregation pheromones of common tropical orchard pests. It is equipped with a high-precision liquid level sensor and an electrically controlled precision metering valve, automatically adjusting the release dosage according to the pest species and environmental conditions, and achieving intelligent and precise release in conjunction with a temperature control switch. The insect-attracting light strip uses a specific wavelength LED light source, optimized for the spectral sensitivity of common tropical orchard pests, and has an intelligent brightness adjustment function, automatically adjusting the light brightness according to the ambient light intensity. It works in conjunction with the pheromone releaser to form a dual trapping mechanism of "light attraction + chemical attraction". The stepper motor drives the rollers to rotate, realizing the periodic replacement of the sticky surface.

[0108] See Figure 3The pest information holographic collection system includes a high-resolution optical imaging module, an edge computing processing module, and a multi-channel pheromone intelligent release module. The high-resolution optical imaging module is equipped with a high-resolution CMOS image sensor with a pixel resolution of no less than 8 million pixels. It features macro shooting, autofocus, and optical image stabilization to ensure clear capture of pest details. It integrates multi-sensor data fusion technology, using a Kalman filter algorithm to achieve real-time fusion processing of multi-sensor data. The edge computing processing module integrates an ARM architecture processor, a 4G wireless communication module, a Beidou positioning module, and an intelligent temperature control and heat dissipation module. It is responsible for real-time processing, storage management, and system coordination control of image data, achieving millisecond-level real-time monitoring through 4G+WiFi6 dual-mode communication technology. The multi-channel pheromone intelligent release module uses microfluidic chip technology to achieve precise regulation and targeted release of pheromones from different pest species. It supports specific pheromone formulations for 15 common tropical orchard pests and is equipped with a high-sensitivity temperature control switch to automatically control the pheromone release intensity according to changes in ambient temperature.

[0109] Furthermore, the multi-model fusion AI recognition system adopts a multi-layered architecture, including a YOLOv11 target detection model, a self-supervised learning model, an intelligent dialogue system, and a biomathematical prediction model. The YOLOv11 target detection model uses the latest YOLOv11 architecture, including a Backbone network, a Neck network, and a Head network. Through deep learning technology, it achieves precise pest location and species identification with an accuracy rate of no less than 98%, and the time from image acquisition to output of the recognition result is only 0.05 seconds. The self-supervised learning model employs five innovative tasks: rotation prediction, contrastive learning, mask reconstruction, color restoration, and texture analysis, significantly reducing the need for labeled data and significantly improving recognition accuracy. The intelligent dialogue system supports natural language interaction, allowing users to communicate with the system via voice or text to obtain professional pest control advice. The biomathematical prediction model establishes a multi-dimensional pest population prediction model based on entomological principles, including a temperature adaptation model, a humidity adaptation model, a pheromone efficacy model, and a population dynamics model. These four models work synergistically to support medium- and long-term predictions with an accuracy rate of no less than 90%.

[0110] Furthermore, the solar photovoltaic power supply system includes solar photovoltaic panels, a solar voltage regulator, and a smart circuit breaker. The solar photovoltaic panels utilize high-efficiency monocrystalline silicon solar cells with a conversion efficiency of no less than 20%, and are equipped with automatic sun tracking to maximize solar energy utilization. The solar voltage regulator integrates MPPT (Maximum Power Point Tracking) technology, using intelligent algorithms to track the maximum power point of the solar panels in real time, ensuring efficient energy conversion and stable output, and providing overvoltage protection, short-circuit protection, and lightning protection. The smart circuit breaker provides overload and short-circuit protection to ensure system electrical safety and prevent equipment damage due to electrical faults. The entire power supply system generates electricity through solar photovoltaic panels, stabilizes the output voltage through the solar voltage regulator, and finally protects the circuit with a smart circuit breaker to achieve stable power supply.

[0111] Furthermore, the predictive model software system includes a front-end user interface (UI) and a back-end model and database. The front-end UI is developed based on Web technology and supports access from multiple terminals. The back-end model and database integrate time series analysis and machine learning algorithms to establish a predictive model for pest population changes. It is equipped with a high-performance database system to store historical monitoring data and analysis results, and adopts a hybrid architecture of MySQL and MongoDB, with MySQL storing structured data and MongoDB storing unstructured data.

[0112] Furthermore, the human-machine interface includes a real-time monitoring interface and a remote control interface. The real-time monitoring interface adopts a visual data display design, including a dynamic pest species distribution map, real-time quantity statistics charts, an environmental parameter monitoring panel, and a device operating status display module, supporting simultaneous monitoring of multiple devices. The remote control interface is used for emergency system shutdown and parameter adjustment in case of emergencies, including functions such as pheromone release control, shooting frequency adjustment, and tape running speed adjustment.

[0113] See Figures 1-9 The working method of the intelligent monitoring and prediction system for tropical orchard pests in this embodiment includes the following steps: Step 1: The system supports both automatic and remote control modes. Before the monitoring work begins, the system can be configured with parameters and adjusted in remote control mode.

[0114] Step 2: When the system is running, the pheromone releaser automatically adjusts the pheromone release intensity according to the ambient temperature and the type of pest. The insect-attracting light strip automatically turns on or off according to the ambient light intensity. The triangular roller insect-trapping module achieves the coordinated operation of the tape and pheromone trapping under the drive of the stepper motor, and the pests are adhered and captured by the sticky tape.

[0115] Step 3: The high-resolution optical imaging module of the system collects pest information data in the work area. The pest information data includes pest species, quantity, location distribution, environmental parameters, pheromone response data, and detection time.

[0116] Step 4: The edge computing processing module processes the types and quantities of pests to obtain a pest distribution map. Combined with the image information obtained by the high-resolution optical imaging module, the distribution and specific locations of the pests are obtained.

[0117] Step 5: Send the pest information data to the multi-model fusion AI recognition system. This system processes the data to obtain pest identification results, pest distribution information, and basic prediction data. Combining the multi-model fusion AI algorithm, the pest distribution map is processed to obtain the ideal pest identification result, and the specific locations of the pests are processed to obtain the ideal pest distribution map. Both are then transmitted to the prediction model software system.

[0118] Step Six: The predictive model software system calculates the corresponding prediction results based on the obtained data, and outputs information such as prevention and control suggestions, risk assessments, and trend charts to the human-computer interaction interface to guide the user in making pest management decisions.

[0119] Step 7: The prediction model software system is equipped with a memory, which is used to store the data acquired by the high-resolution optical imaging module, mainly the pest distribution map model established based on pest information data. In this way, on the one hand, it can be used to make a comprehensive and multi-faceted evaluation of the monitoring work after prediction, and on the other hand, if the system is equipped with a special optimization prediction algorithm, it can be used to optimize the next prediction.

[0120] Step 8: During operation, the system transmits all pest information data, as well as the system's working environment, local weather conditions, weather warnings, and future weather changes, to the human-machine interface via remote transmission modules (4G module, WiFi module, etc.) so that the operator can make the next management plan.

[0121] The above are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the one described above. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A smart monitoring and prediction system for tropical orchard pests, characterized in that, include: Pest trapping module, pest information holographic collection system, multi-model fusion AI recognition system, solar photovoltaic power supply system, and human-computer interaction interface; The pest trapping module includes a triangular roller insect trapping module (6), a pheromone releaser (7), an insect-attracting light strip, and a stepper motor (2); the pest information holographic collection system includes a microscope camera (5), a multi-functional motherboard, and an edge computing processing module; the multi-model fusion AI recognition system includes a YOLO recognition model, a self-supervised learning model, an intelligent dialogue system, and a biomathematical prediction model; the solar photovoltaic smart power supply system includes a solar photovoltaic panel (1), a solar voltage regulator, and an intelligent circuit breaker; the human-machine interface includes a real-time monitoring interface and a remote control interface.

2. The intelligent monitoring and prediction system for tropical orchard pests according to claim 1, characterized in that: The lower roller of the triangular roller insect-catching module supplies insect-catching tape, the middle roller attracts insects, and the upper roller retrieves them. The three rollers are arranged in an equilateral triangle and are fixed at the top and bottom by the left and right side shaft support brackets (3). The middle roller is fixed to the sheet metal shell (4). The roller surface of the triangular roller insect-catching module is covered with sticky insect tape. The stepper motor (2) drives the roller to rotate, realizing the periodic replacement of the sticky insect surface. The insect-attracting lamp (7) is equipped with a photosensitive switch to realize automatic light control.

3. The intelligent monitoring and prediction system for tropical orchard pests according to claim 1, characterized in that: The microscope camera (5) is equipped with a high-resolution CMOS image sensor with a pixel resolution of no less than 8 million pixels. It has macro shooting and autofocus functions and can capture high-definition images of tiny pests. The multi-functional motherboard module integrates an ARM architecture processor, a 4G wireless communication module, a Beidou positioning module and a high-efficiency heat dissipation module. It is responsible for the preliminary processing, storage management and system coordination control of image data to ensure the stability and accurate positioning of the device in remote operation. The edge computing processing module is deployed on the Jetson Nano embedded platform and is used to complete the preprocessing and feature extraction of pest images on the device side.

4. The intelligent monitoring and prediction system for tropical orchard pests according to claim 1, characterized in that: The YOLO recognition model adopts the YOLOv11 architecture, whose core structure includes: an improved CSPNet (Cross Stage Partial Network) as the backbone for feature extraction, a PathAggregation Network for multi-scale feature fusion, and a decoupled head for final detection, which identifies the types and quantities of pests on the sticky tape in real time; the self-supervised learning model adopts four tasks: contrastive learning, mask reconstruction, color restoration, and texture analysis; the biomathematical prediction model is based on temperature adaptation equation, humidity adaptation equation, pheromone efficacy equation, and population dynamics equation, combined with historical pest data and environmental parameters, to output the pest outbreak risk level for the next 7 days.

5. The intelligent monitoring and prediction system for tropical orchard pests according to claim 1, characterized in that: The solar photovoltaic power supply system includes a solar photovoltaic panel (1), a solar voltage regulator and an intelligent circuit breaker; the solar photovoltaic panel adopts a high-efficiency monocrystalline silicon solar cell panel and is equipped with AI-driven intelligent light tracking + temperature tracking dual-mode function; the solar voltage regulator integrates fourth-generation MPPT maximum power point tracking technology; the intelligent circuit breaker provides triple overload protection and instantaneous short circuit protection functions.

6. The intelligent monitoring and prediction system for tropical orchard pests according to claim 1, characterized in that: The remote control interface of the intelligent human-computer interaction interface supports mobile access, and users can adjust the pheromone release strategy and view warning information through voice input on this interface.

Images