Crop insect pest image intelligent identification system based on electronic information technology

Abstract

The invention discloses a crop pest image intelligent identification system based on an electronic information technology, and relates to the technical field of agricultural informatization. The system comprises an image acquisition module which acquires RGB image data of a target crop area and synchronously acquires associated environment monitoring data and image shooting time data; the preprocessing module preprocesses the data; the insect pest recognition module outputs insect pest type data and insect body position frame selection coordinates through an improved YOLO recognition model; the comparison module generates insect pest development state data; a risk assessment module calls a pest hazard level dictionary matching risk coefficient, and calculates an environmental suitability value in combination with environmental monitoring processing data; the risk scoring module generates a comprehensive risk score; and the early warning decision module matches the prevention and treatment scheme data when the comprehensive risk score exceeds a preset threshold. According to the technology, intelligence and precision of insect pest recognition and early warning are realized, and the agricultural insect pest prevention and control efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of agricultural information technology, and in particular to an intelligent crop pest image recognition system based on electronic information technology. Background Art

[0002] Traditional crop pest monitoring relies primarily on a combination of manual inspections and simple sensing equipment, which presents significant technical limitations. Inspectors must conduct in-depth field surveys, inspecting each area and determining the presence of pests by visually observing plant abnormalities or directly searching for insects. This method is not only time-consuming and labor-intensive, but also heavily influenced by subjective experience. Minor infestation characteristics or hidden pests can easily be overlooked, while environmental disturbances (such as natural leaf damage) often lead to misjudgments. Furthermore, manual inspections cannot achieve continuous, dynamic monitoring. Subtle signs of an early outbreak are often missed due to gaps in inspections, often leading to irreversible damage by the time they are discovered.

[0003] While existing automated detection methods have attempted to incorporate image recognition technology, they remain limited to single-source visual data analysis. Conventional systems capture crop images solely through cameras and employ fixed thresholds or simple classification models to identify pests. This completely ignores the dynamic impact of environmental factors on pest behavior. For example, they ignore the accelerating effects of temperature and humidity changes on pest reproduction and fail to consider how lighting conditions alter insect activity patterns, resulting in significant discrepancies between identification results and the actual extent of damage. Furthermore, such systems generally lack the ability to analyze pest evolution trends, providing only static assessments of pest presence and absence, unable to predict insect swarm spread trends or outbreak risk levels. Disrupted data chains further exacerbate decision-making lags: image recognition, environmental monitoring, and historical records operate in isolation, creating fragmented information silos. This results in prevention and control decisions relying on fragmented information, making it difficult to generate scientific risk assessments and timely response plans.

[0004] In response to the above problems, this field urgently needs to build an intelligent recognition system that integrates multi-source dynamic data, correlates environmental and biological behavior patterns, and has the ability to continuously learn and optimize, so as to break through the technical bottlenecks of traditional monitoring models in timeliness, accuracy and predictive ability. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the present invention provides an intelligent crop pest image recognition system based on electronic information technology.

[0006] In order to achieve the above object, the technical solution of the present invention is as follows:

[0007] The present invention discloses an intelligent crop pest image recognition system based on electronic information technology, comprising:

[0008] An image acquisition module is configured to acquire RGB image data of a target crop area and simultaneously acquire environmental monitoring data and image capture time data associated with the RGB image data; a preprocessing module is configured to preprocess the RGB image data, environmental monitoring data, and image capture time data to generate a processed image matrix, environmental monitoring processing data, and image capture time processing data; and a pest identification module is configured to fuse the image matrix with the environmental monitoring processing data using an improved YOLO recognition model to output pest type data and insect body location frame selection coordinates.

[0009] a comparison module for comparing the image shooting time processing data with pest evolution stage data in a pre-stored historical pest spatiotemporal database to generate pest development status data;

[0010] A risk assessment module is used to call a preset pest hazard level dictionary to obtain a risk coefficient matching the pest type data, and calculate an environmental suitability value in combination with the environmental monitoring and processing data;

[0011] A risk scoring module is used to collect insect population density data based on the insect body position frame coordinates, and to generate a comprehensive risk score by integrating the insect pest development status data, the risk coefficient, the environmental suitability value, and the insect population density data;

[0012] The early warning decision module is used to call the pre-stored prevention and control strategy database to match the corresponding prevention and control plan data according to the pest type data when the comprehensive risk score exceeds the preset risk threshold.

[0013] Compared with the prior art, the present invention has the following beneficial effects:

[0014] 1. By fusing environmental monitoring data with image matrices within an improved YOLO recognition model, this system transcends the reliance of traditional visual recognition models on isolated image data, significantly enhancing the ability to identify pest characteristics in complex farmland scenarios. A dynamic weighting mechanism for environmental factors enables the model to autonomously emphasize key regional features strongly correlated with pest behavior (such as egg distribution in high-temperature and high-humidity environments), while simultaneously mitigating interference noise such as lighting variations and leaf occlusion, thereby reducing false positives caused by environmental interference. This fusion mechanism simultaneously correlates pest physiological characteristics with environmental adaptability, ensuring that identification results more closely reflect actual pest activity.

[0015] 2. This invention leverages the collaborative analysis capabilities of multi-source dynamic data to systematically construct a scientific and quantitative risk assessment model. Pest development data accurately identifies the stage of insect spread by comparing historical evolution patterns, while environmental suitability values ​​reflect the driving force behind pest reproduction in real time. Together with insect population density data, these two form a three-dimensional evaluation system, fundamentally transforming the traditional static threshold judgment model. This coupled mechanism enables a comprehensive risk score to simultaneously characterize both the immediate intensity of the hazard and the potential for an outbreak, providing a timely and predictive scientific basis for prevention and control decisions, avoiding the delays and resource waste caused by single indicators.

[0016] 3. This system, leveraging a closed-loop, modular architecture, seamlessly transitions from risk identification to prevention and control decision-making. When a comprehensive risk score triggers an early warning mechanism, the prevention and control strategy database automatically matches a multi-level response plan based on the type of pest, ensuring that prevention and control measures are fully aligned with the real-time pest situation and environmental carrying capacity. This system offers greater practical value than traditional systems that simply indicate the presence of pests, meeting modern agriculture's demand for real-time monitoring and precise prevention of rice field pests. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The disclosure of the present invention is described with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. In the drawings, the same reference numerals are used to refer to the same components. Among them:

[0018] Figure 1 A module connection diagram of the system of the present invention;

[0019] Figure 2 A flowchart of the steps of the modules of the system of the present invention;

[0020] Figure 3 Module connection diagram of the closed-loop optimization module of the present invention;

[0021] Figure 4 Module connection diagram of the instruction output module of the present invention;

[0022] Figure 5 This is a module connection diagram of the user feedback module of the present invention. DETAILED DESCRIPTION

[0023] It is easy to understand that according to the technical solution of the present invention, without changing the essential spirit of the present invention, a person skilled in the art can propose a variety of interchangeable structural modes and implementation modes. Therefore, the following specific embodiments and drawings are only exemplary descriptions of the technical solution of the present invention and should not be regarded as the entire invention or as a limitation or restriction of the technical solution of the present invention.

[0024] Application Overview:

[0025] Existing technologies primarily rely on manual inspections or single visual recognition models, making it difficult to balance recognition accuracy and risk prediction capabilities in complex environments. Traditional methods, in pest-prone environments like high temperature and high humidity, ignore the dynamic impact of environmental factors on pest behavior, resulting in identification results that deviate from the actual level of damage. Furthermore, static insect population counts cannot correlate with insect swarm evolution trends, causing prevention and control decisions to lag behind the pest outbreak cycle. Existing systems generally lack mechanisms for collaborative analysis of multi-source data. Environmental monitoring, image recognition, and historical records operate in isolation, forming fragmented information silos that make it difficult to generate scientific and quantitative risk assessment models.

[0026] To address these issues, the inventors discovered a strong correlation between environmental factors (temperature, humidity, and light) and pest activity patterns, a nonlinear critical point in pest population growth curves, and that historical pest data reveals patterns of spatiotemporal evolution. By establishing a coupled environmental-organismal behavior model, they proposed a dynamic weighting and multi-dimensional scoring mechanism: increasing environmental suitability strengthens the weight of biological response characteristics, triggering escalated control measures when pest population density exceeds an inflection point, and leveraging historical evolutionary patterns to predict spread.

[0027] Specifically, the system simultaneously collects farmland images and environmental data. Using an improved YOLO model, it maps environmental parameters into an image weight matrix, enhancing regional responses to pest signatures. A comparison module links real-time data with a historical spatiotemporal database of pests to output quantitative pest development status values. A risk assessment module integrates pest population density, environmental suitability, risk factors, and pest development status to construct a three-dimensional dynamic scoring model. When the combined risk score exceeds a threshold, the control strategy database automatically matches solutions based on environmental suitability priorities and adds additional response levels as pest population density continues to increase. A closed-loop optimization module continuously updates the pest evolution model and control strategy library based on user feedback.

[0028] Through the above technical solution, this application effectively overcomes the recognition bias caused by environmental interference and the decision lag of static assessment. The environmental weighted fusion mechanism enhances the robustness of feature recognition, the three-dimensional scoring model simultaneously quantifies immediate harm and potential risks, and the closed-loop optimization system ensures that the strategy continuously adapts to actual needs. This method provides reliable technical support for the precise prevention and control of crop pests and is particularly suitable for large-scale farmland monitoring scenarios where multiple environmental factors are intertwined.

[0029] After introducing the basic concept of the present invention, embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0030] Example:

[0031] like Figure 1 As shown, an intelligent crop pest image recognition system based on electronic information technology includes:

[0032] An image acquisition module is used to acquire RGB image data of the target crop area and simultaneously acquire environmental monitoring data and image capture time data associated with the RGB image data;

[0033] The preprocessing module is used to preprocess the RGB image data, environmental monitoring data, and image shooting time data to generate the processed image matrix, environmental monitoring processing data, and image shooting time processing data; the pest identification module is used to fuse the image matrix with the environmental monitoring processing data through the improved YOLO recognition model, and output the pest type data and the insect body location frame selection coordinates;

[0034] A comparison module is used to compare the image shooting time processing data with the pest evolution stage data in the pre-stored historical pest spatiotemporal database to generate pest development status data;

[0035] The risk assessment module is used to call the preset pest hazard level dictionary to obtain the risk coefficient matching the pest type data, and calculate the environmental suitability value in combination with the environmental monitoring data;

[0036] The risk scoring module is used to collect insect population density data based on the coordinates of the insect body position frame, and integrate the pest development status data, risk coefficient, environmental suitability value and insect population density data to generate a comprehensive risk score;

[0037] The early warning decision module is used to call the pre-stored prevention and control strategy database to match the corresponding prevention and control plan data according to the pest type data when the comprehensive risk score exceeds the preset risk threshold.

[0038] The image acquisition module refers to a device that collaboratively collects multi-dimensional data through a multispectral camera and an unmanned aerial vehicle (UAV). This can be achieved by combining a fixed camera array with an UAV flight platform to simultaneously acquire visible light images and their associated environmental parameters. The preprocessing module standardizes raw data, using an image scaling algorithm and environmental data alignment methods to eliminate data noise and unify the input format. The improved YOLO recognition model is a target detection algorithm that incorporates environmental factor weights. It uses convolution kernel mapping to achieve feature fusion between the image matrix and environmental data, improving the accuracy of pest identification. Pest development status data refers to an evolutionary trend indicator generated through time series comparison. This can be achieved by similarity matching spatiotemporal labels in a historical database to predict the spread of insect swarms. The comprehensive risk score is a dynamic assessment that integrates the pest development status, environmental suitability, insect population density, and risk factor. This can be achieved through a weighted calculation formula to quantify the current pest threat level.

[0039] like Figure 2Specifically, during operation, the system uses a fixed camera array to capture visible light images of farmland at a preset frequency. It also uses drones to capture multi-angle images, simultaneously recording environmental parameters such as temperature, humidity, and light intensity, as well as GPS positioning information. The preprocessing module normalizes the image size, performs illumination compensation, and performs denoising. It then aligns the environmental data with timestamps to generate standardized input data. The pest identification module fuses the processed image matrix with the environmental data, using an improved YOLO model to identify pest species and mark their location coordinates. The comparison module matches current time data with historical pest evolution stages to generate pest development trend forecasts. The risk assessment module queries a preset hazard level dictionary based on pest species and calculates a suitability score based on environmental parameters. The risk scoring module counts the number of insects per unit area and integrates multi-dimensional data to generate a dynamic risk score. When the score exceeds a threshold, the early warning and decision-making module automatically calls the matching control plan database to output a response strategy.

[0040] Through the above technical solutions, this application effectively solves the problems of data isolation, high misjudgment rate and delayed decision-making in traditional monitoring methods. The multi-source data fusion mechanism improves the environmental adaptability of pest identification and avoids feature misextraction due to changes in lighting conditions. The dynamic risk assessment model integrates insect population density, environmental suitability and historical evolution data to accurately quantify the threat level of pests. The early warning decision-making module achieves precise matching of prevention and control strategies, such as automatically triggering biological control plans in the early stages of pests to prevent the expansion of insect swarms. The closed-loop data flow design ensures continuous optimization of the system, such as automatically updating model parameters after correcting the recognition results through user feedback, and gradually improving the recognition accuracy.

[0041] The present application further proposes that the image acquisition module includes:

[0042] A fixed multispectral camera array is deployed at a preset height in the farmland to collect RGB image data in the visible light band at a first sampling frequency;

[0043] The drone is called to collect multi-angle images at a second sampling frequency along a preset flight path to obtain GPS positioning data, environmental monitoring data and image shooting time data associated with RGB image data; wherein the environmental monitoring data includes temperature, humidity, and light intensity.

[0044] Among them, a fixed multispectral camera array refers to a cluster of devices composed of multiple cameras with visible light band imaging capabilities. Specifically, it can be implemented using a wide-angle camera array installed on a farmland bracket. By deploying it at a preset height, it can cover a larger range of crop areas, solving the problem of limited viewing angle of a traditional single camera.

[0045] The first sampling frequency refers to the time interval for the fixed camera to capture images. Specifically, a timing trigger mechanism can be adopted, such as taking a picture every 30 minutes, to ensure continuous monitoring of the pest status in the farmland. The preset flight path of the drone refers to the pre-planned disordered or ordered flight trajectory. Specifically, an automatic route planning algorithm using GPS navigation can be adopted to capture images from multiple angles to compensate for the monitoring blind spots of the fixed camera.

[0046] The second sampling frequency refers to the period for the drone to perform the image capture task. Specifically, a time interval setting complementary to that of the fixed camera can be adopted, such as conducting a full-area scan every 2 hours, to improve the timeliness of dynamic pest monitoring. The GPS positioning data refers to the geographical location information recorded when the drone takes pictures. Specifically, a high-precision differential GPS module can be adopted to provide a spatial coordinate reference for subsequent tracing of the pest location.

[0047] Specifically, a fixed multi-spectral camera array is deployed on the pre-set brackets or poles in the farmland, and its installation height can be within the range of 3 meters to 5 meters, covering the crop area below through a wide-angle lens. The camera array periodically collects RGB image data at the first sampling frequency, such as triggering a shot every 30 minutes, to form a continuous monitoring data stream. At the same time, the drone follows the preset flight path, such as covering the entire farmland area in a "field" - shaped trajectory, and executes the flight task at the second sampling frequency. During the flight, the camera carried by the drone takes pictures of the crops from different angles, and synchronously records environmental parameters such as GPS positioning data, temperature, humidity, and light intensity at the time of shooting. The collaborative work of the fixed camera and the drone forms a multi-level data acquisition network. Among them, the fixed camera provides high-frequency basic monitoring data, and the drone supplements multi-angle images and spatial position information. The combination of the two can effectively solve the monitoring blind spot problem caused by traditional single data sources.

[0048] Through the above technical solutions, this application can achieve full-area coverage of farmland pest monitoring and synchronous acquisition of multi-dimensional data. The high-frequency shooting of the fixed camera captures the temporal variation law of pest occurrence. The multi-angle images of the drone enhance the recognition probability of hidden pests. The synchronous recording of environmental data and spatial coordinates provides key inputs for subsequent analysis of pest spread. This solution effectively solves the monitoring blind spot problem caused by single data sources in traditional methods, and at the same time avoids the technical defects of low efficiency of manual inspection and fragmented analysis of environmental factors, providing high-precision and multi-dimensional basic data support for intelligent pest identification. [[ID=lo]]

[0049] This application further proposes to preprocess the RGB image data, environmental monitoring data, and image shooting time data to generate processed image matrices, environmental monitoring processed data, and image shooting time processed data. The specific steps of the preprocessing include:

[0050] The input RGB image data is uniformly scaled to 640×640 pixels using a bilinear interpolation algorithm;

[0051] The CLAHE algorithm is used to perform illumination compensation on RGB image data;

[0052] Apply the non-local means denoising method to remove image noise from RGB image data;

[0053] The processed RGB image data is aligned with the GPS positioning data, environmental monitoring data and image capture time data through timestamps;

[0054] Output processed image matrix, GPS positioning processing data, environmental monitoring processing data and image shooting time processing data; the processed data is a tensor data structure that meets the input requirements of the improved YOLO recognition model.

[0055] Bilinear interpolation is an image scaling method that calculates new pixel values ​​based on a weighted average of adjacent pixels. It can be implemented using linear interpolation between pixels and is used to maintain edge smoothness and detail integrity when resizing an image. CLAHE is a contrast-constrained adaptive histogram equalization method. It uses block-wise histogram equalization and contrast limiting to eliminate local overexposure or underexposure caused by uneven lighting. Non-local means denoising uses non-locally similar regions within an image to suppress noise. It can be implemented using pixel block similarity calculations and weighted superposition to remove random noise interference while preserving texture detail. Timestamp alignment is the synchronization of time series from multiple sources. It can be implemented using time-coded parsing and data frame matching algorithms to ensure temporal consistency between image data and associated environmental data. Tensor data structures are multidimensional arrays of data. They can be constructed using a four-dimensional matrix (batch, height, width, channels) to meet the standardized input data requirements of deep learning models.

[0056] Specifically, the preprocessing process first normalizes the original RGB image size. Input images of varying resolutions are then scaled to a fixed size, such as 640×640 pixels, using a bilinear interpolation algorithm to eliminate image scale distortion caused by device variations. The scaled image is then illuminated, and adaptive histogram equalization (CLAHE) is performed on each color channel to limit the extent of local contrast enhancement and avoid excessive noise enhancement. Non-local means denoising is then applied, calculating pixel weights based on global image similarity. This method suppresses high-frequency noise while preserving leaf texture and insect outlines. After image processing, the processed image matrix is ​​timestamped with GPS positioning data and environmental monitoring data to ensure a strict temporal match between the multi-source data collected at the same time. Finally, the image matrix is ​​converted into a four-dimensional tensor structure, and the environmental monitoring data and time data are encoded as additional feature vectors, forming a standardized data package that meets the model input requirements.

[0057] Through the above technical solutions, this application effectively solves the problem of recognition errors caused by environmental interference in image quality. It improves the recognition of insect outlines through illumination compensation and noise suppression, eliminates time misalignment in multi-source data through timestamp alignment, and standardizes tensor structures to ensure consistency in model input. This reduces the false detection rate and missed detection rate of subsequent recognition modules, providing reliable data preprocessing support for pest species identification and location positioning.

[0058] This application further proposes that the image matrix and environmental monitoring data are integrated through an improved YOLO recognition model to output pest species data and insect body location frame selection coordinates. The specific steps include:

[0059] The environmental monitoring processed data is mapped into a weight matrix of the same dimension as the image matrix through a 1×1 convolution kernel, and multiplied element-by-element with the RGB channels of the image matrix;

[0060] The current image matrix is ​​associated with the pre-stored pest time series sample library, and the cosine similarity algorithm is used to compare the historical pest features of the previous three frames and perform confidence matching, and the confidence of the pest evolution stage is output;

[0061] The image matrix is ​​scored for clarity and illumination uniformity to obtain image quality. When the image quality is lower than the quality threshold, the image acquisition module is triggered to reshoot, and the quality threshold is dynamically adjusted by weighted calculation of image quality.

[0062] The improved YOLO recognition model includes:

[0063] Environmental feature injection channel: Add an environmental data fully connected layer at the input of the Backbone network to convert environmental monitoring data into a 256-dimensional feature vector;

[0064] Spatiotemporal attention mechanism: The spatiotemporal attention module is introduced in the Neck network layer, and the formula for outputting the confidence of the pest evolution stage is:

[0065]

[0066]

[0067] Parameter Description: ; : Historical pest feature vector (i=1,2,3 corresponds to the first three frames); : The cosine similarity between the current image and the historical features of the i-th frame;

[0068] The cosine similarity algorithm is used to quantify the similarity between the current pest characteristics and the historical characteristics. The comprehensive confidence is calculated by averaging the similarity of the historical characteristics of the previous three frames. The higher the confidence, the stronger the continuity of the pest development and the higher the probability of being in the same evolutionary stage.

[0069] Dynamic threshold adjustment unit: linearly adjusts the detection quality threshold according to the light intensity value. The quality threshold adjustment formula is:

[0070]

[0071] in, : Basic quality threshold (preset constant); I: Current ambient light intensity (obtained by sensor); : Preset light reference value (set according to crop light requirements).

[0072] Basic threshold It increases linearly with the increase of light intensity I. When the light intensity is higher than the baseline value, ), the system automatically increases the quality requirements to avoid recognition errors caused by overexposed images in strong light environments, and appropriately relaxes the requirements in low light environments.

[0073] Specifically, environmental monitoring data is convolutionally computed to generate a weight matrix, which is then fused with image features, enabling the model to automatically focus on environmentally sensitive areas during recognition. Historical pest signatures are compared for similarity using a time series sample library, and a confidence matching mechanism is incorporated to enhance the continuity of pest evolution trends. When the image quality score falls below a dynamic threshold, the system automatically triggers a data re-collection process to prevent low-quality input from impacting recognition accuracy.

[0074] Through the above technical solution, this application effectively solves the recognition bias problem caused by the separation of environmental factors and visual features, enhances the ability to continuously judge the evolution trend of pests, ensures the reliability of input data through a dynamic quality assessment mechanism, and significantly improves the accuracy of pest identification and system robustness.

[0075] The present application further proposes to call a preset pest hazard level dictionary to match the risk coefficient of the pest type data, and combine it with the current environmental monitoring data to calculate the environmental suitability value. The calculation method of the environmental suitability value is:

[0076] Query the optimal range of target pests corresponding to the pest species data from the preset environmental suitability comparison table , , ;

[0077] The suitability score of a single environmental factor is calculated using a piecewise function. The formula is as follows:

[0078] ;

[0079] in, is the actual monitoring value of the environmental factor, is the median value of the optimal range, is the optimal range width;

[0080] The weighted sum generates the environmental suitability value:

[0081] ;

[0082] Among them, the environmental suitability value is the total score, 、 、 They are the temperature factor single score, humidity factor single score, light factor single score, and environmental factor weight Obtained through training on historical pest outbreak data.

[0083] Among them, the environmental suitability comparison table refers to a database that stores the range of environmental parameters required for the survival and reproduction of various types of pests. Specifically, it can be implemented using a relational database table structure to establish a mapping relationship between the types of pests and the corresponding temperature, humidity, and light intensity threshold intervals. The piecewise function refers to a mathematical relationship that calculates the score based on the degree of deviation between the environmental monitoring value and the optimal range of the pest. Specifically, it can be implemented using a conditional judgment statement to quantify the promoting or inhibiting effect of environmental factors on the survival of pests. Weighted summation refers to a calculation method that weights the importance of the scores of different environmental factors. Specifically, it can be implemented using matrix multiplication to reflect the differences in the weights of the impact of different environmental factors on the development of pests.

[0084] The weight coefficient ( ) Through closed-loop training optimization of historical pest outbreak data, for example, when the system detects rice planthopper pests, it will automatically load the preset weight set , precisely adapted to its high humidity sensitivity characteristics.

[0085] Specifically, once the pest species data is identified, the system extracts the pest's optimal temperature, humidity, and light range from the environmental suitability comparison table. For example, if a pest's optimal temperature range is 20-30°C, the median temperature factor is 25°C, and the range width is 5°C. When the actual monitored temperature is 28°C, the temperature factor score is calculated as 0.8 using a piecewise function. Each environmental factor score is multiplied by the weight coefficient obtained through training, such as a temperature weight of 0.4, humidity 0.3, and light 0.3. The weighted sum constitutes the environmental suitability value. This score dynamically reflects the extent to which the current environment promotes pest reproduction, providing a quantitative basis for risk assessment.

[0086] Through the above technical solution, this application can accurately quantify the role of environmental conditions in promoting the development of insect pests, combine the physiological characteristics of pests with historical outbreak patterns, and dynamically generate environmental suitability values ​​that reflect actual risks, providing accurate quantitative input for subsequent comprehensive risk assessments, and effectively avoiding prevention and control decision-making errors caused by deviations in environmental factor assessments.

[0087] The present application further proposes to collect insect population density data based on insect body position frame coordinates, and integrate pest development status data, risk coefficient, environmental suitability value and insect population density data to generate a comprehensive risk score. The generation of the comprehensive risk score specifically includes:

[0088] Establish a preset pest hazard level dictionary to store the basic risk factors of various pests in the form of key-value pairs ;

[0089] According to the pest type data output by the pest identification module, the pest hazard level dictionary is searched and the basic risk coefficient is used to Get the corresponding risk factor ;

[0090] Based on the insect body position, select the coordinates and count the number of insects per unit area to obtain the insect population density data. ;

[0091] The final comprehensive risk score fusion formula is obtained, and its calculation formula is as follows:

[0092] ;

[0093] in is the preset weight coefficient, satisfying ; is the environmental suitability value; This is the pest development status data. The weight coefficient is stored in the system configuration database and the preset value is called according to the crop type.

[0094] Among them, the pest hazard level dictionary refers to a database that stores the basic risk coefficients of various types of pests. It can be implemented using a hash table structure, using the pest species name as the key and the basic risk coefficient as the value for storing, and is used to quickly query the initial risk value of a specific pest.

[0095]

[0096] Insect population density data refers to the statistical value of the number of insects per unit area, which can be achieved by clustering the insect position coordinates and calculating the number of insects per unit area, reflecting the spatial distribution density of insect pests. , statistics of insect population density per unit area , the formula is as follows:

[0097]

[0098] in, : The number of insect bodies identified (the number of selected coordinates); : Effective monitoring area (square meters), the calculation formula is:

[0099]

[0100] in, , is the image resolution (pixels), is the image resolution (m / pixel), k is the crop coverage correction factor (the default value is 0.6-0.9), and the crop coverage correction factor k is introduced to solve the area calculation deviation caused by the difference in crop planting density in the image.

[0101] The comprehensive risk score fusion formula refers to a mathematical calculation model that integrates multi-dimensional influencing factors. It can be implemented by linear weighted summation, balancing the contribution ratios of environmental suitability, insect population density and pest development status through preset weight coefficients.

[0102] The comprehensive risk score fusion formula uses logarithmic transformation Prevent high-density values ​​from dominating the score; weight coefficient Stored in the system configuration database, preset configuration sets can be loaded for different crop types (such as rice weight set, wheat weight set).

[0103] Specifically, in the process of generating a comprehensive risk score, the basic risk coefficient corresponding to the current pest type is first obtained through the pest hazard level dictionary. This coefficient reflects the inherent degree of harm of a specific pest. Then, based on the insect body detection results, the insect population density per unit area is calculated to quantify the current scale of the pest. The environmental suitability value represents the degree to which the current environmental conditions promote pest reproduction, and the pest development status data reflects the evolution trend of the pest. These three dynamic factors are weighted and fused with the basic risk coefficient through preset weight coefficients to form a comprehensive risk score. For example, when the environmental suitability value is high and the insect population density continues to increase, even if the basic risk coefficient is low, the comprehensive score may still exceed the threshold and trigger an early warning. The weight coefficient can be adjusted according to different crop types. For example, the weight of environmental factors can be increased in greenhouse planting scenarios, while in open-air farmland, the focus can be on changes in insect population density.

[0104] Through the above technical solutions, this application effectively solves the problems of single data dimension and inaccurate static threshold judgment in traditional pest risk assessment. By dynamically integrating multi-source data such as environmental conditions, insect swarm size and evolutionary trends, it can accurately identify high-risk pest outbreak scenarios and provide a quantitative basis for prevention and control decisions. For example, in a high temperature and high humidity environment, even if the insect population density does not reach the traditional threshold, an early warning can still be triggered due to the high environmental suitability value, avoiding losses caused by the environment accelerating the development of pests. At the same time, the configurability of the weight coefficient enables the system to adapt to the risk assessment needs of different crop types and planting environments.

[0105] The present application further proposes that when the comprehensive risk score exceeds a preset risk threshold, the pre-stored control strategy database is called based on the pest type data to match the corresponding control plan data. The control plan data matching process also includes:

[0106] When it is detected that the comprehensive risk score of the same area exceeds the preset risk threshold for three consecutive times, an upgraded prevention and control plan will be generated;

[0107] Upgrade the prevention and control plan to set priority sorting rules and output a combination of prevention and control plans with validity periods:

[0108] (1) Current environmental suitability value When >0.8, biological control is preferred;

[0109] (2) After biological control is initiated, if the nymph population density continues to increase, chemical control should be added;

[0110] (3) Current insect population density data of the number of insects per unit area Automatically increase the chemical concentration ratio when the density is greater than the threshold;

[0111] (4) When the historical efficiency of the same solution is less than 60%, start searching for alternative solutions.

[0112] Among them, an upgraded control plan refers to a strategy that dynamically adjusts the intensity of control measures based on multiple consecutive risk warnings. This can be achieved by setting trigger thresholds and multi-condition judgment logic, and is used to address the lack of adaptability of a single control plan to the evolution of complex pests. Among them, biological control refers to an ecological management method that uses natural enemies or microorganisms to suppress pests. Specifically, this can be achieved by releasing parasitic wasps or spraying Bacillus thuringiensis, and is used to prioritize reducing the use of chemical agents when the environment is suitable.

[0113] Chemical concentration scaling dynamically adjusts the application intensity based on pest population density. This is achieved through a preset concentration-density mapping table or linear interpolation calculation to improve pest control efficiency during an outbreak. Density thresholds are preset based on the biological characteristics of the target pests, for example, 50 insects / ㎡ for rice fields and 30 insects / ㎡ for wheat fields.

[0114] Among them, alternative solution retrieval refers to the process of initiating new strategy matching when the historical prevention and control effects do not meet the standards. It can be achieved by searching the prevention and control strategy database through a similarity matching algorithm to avoid waste of resources caused by repeated invalid solutions.

[0115] If a risk warning is triggered three times consecutively within a preset period in the same farmland area, the system determines that the pest has entered a phase of accelerated spread and generates an upgraded control plan comprising multiple measures. During implementation, biological control is first selected as the primary measure based on real-time environmental scores. For example, when temperature and humidity conditions are suitable for the activity of natural enemies, parasitic wasp releases are prioritized. If insect population density continues to rise after biological control, the system automatically adds chemical control measures, such as increasing the spraying frequency on top of the existing application. If the monitored insect population per unit area exceeds a critical threshold (for example, over 50 insects per square meter), the pesticide concentration is increased by a preset ratio to ensure effective control. The system also continuously compares historical control records. If the insect population decline rate after the recent implementation of a similar plan falls below a set threshold, it immediately retrieves alternative options from the database (such as changing the pesticide type or introducing physical control devices), thus forming a dynamically optimized control strategy chain.

[0116] Through the above technical solution, this application can timely identify the risk escalation trend in the middle stage of pest development, suppress the spread of insect swarms through a combination of multi-stage measures, and avoid the recurrence of pests caused by the failure of a single prevention and control method; at the same time, dynamically optimize the pesticide application strategy according to environmental conditions and historical data, reduce the abuse of chemical agents while ensuring the prevention and control effect, and form an environmentally friendly pest control system.

[0117] like Figure 3 As shown, the present application further proposes that an intelligent crop pest image recognition system based on electronic information technology also includes a closed-loop optimization module, which performs the following operations:

[0118] Obtain confirmation data and false alarm marks submitted by the user terminal, where the confirmation data includes the image ID, the actual pest type, and the corrected insect body location frame coordinates;

[0119] Based on the confirmed data, data feedback is performed to obtain new data, incrementally update the historical pest spatiotemporal database, and add new pest samples with spatiotemporal labels; update the pest damage level dictionary and adjust the risk factor based on actual damage records; update the control strategy database and associate the records of the effectiveness of the implementation of the control plan data; the effectiveness rate is the insect population decline rate after the same plan is implemented within 7 days after extraction from the control strategy database;

[0120] Call the newly added data to fine-tune the parameters of the improved YOLO recognition model and update the model parameters of the improved YOLO recognition model.

[0121] Confirmation data refers to user verification of the system's identification results. This can be achieved by using image annotation tools to collect manually corrected pest species and location coordinates, eliminating system misjudgments and supplementing real sample data. Incremental updates refer to the addition of new samples without overwriting existing data. This is achieved by recording new operations in database transaction logs to ensure the timeliness and sample diversity of the historical pest spatiotemporal database. Parameter fine-tuning refers to adjusting model weights based on the newly added data. This is achieved by using transfer learning methods to freeze some network layers and then train the fully connected layers. This is used to improve the model's ability to identify newly emerging pest characteristics.

[0122] Specifically, the closed-loop optimization module receives user-corrected pest data and adds real pest samples to the historical database, forming a pest evolution trajectory with spatiotemporal labels. The pest hazard level dictionary dynamically adjusts the risk factor based on the actual control effectiveness. For example, when the actual damage level of a certain type of pest exceeds the original level, its risk factor weight is automatically increased. The control strategy database calculates the effectiveness of the plan by calculating the rate of decrease in insect population density within seven days, providing data support for subsequent plan optimization. The improved YOLO recognition model uses an incremental learning mechanism. Each parameter fine-tuning only updates the network layer parameters related to the newly added sample features, maintaining the stability of the original recognition capabilities.

[0123] Through the above technical solution, this application solves the problem of model aging caused by data isolation in traditional pest identification systems. By continuously integrating user-corrected data and feedback on control effectiveness, the accuracy of pest identification and the effectiveness of control plans are improved. The closed-loop optimization mechanism enables the system to automatically adapt to the evolution of regional pests and diseases, reducing the risk of misjudgment due to environmental changes or the emergence of new pests, while ensuring the timeliness and targetedness of control strategies.

[0124] like Figure 4As shown, the present application further proposes that a crop pest image intelligent recognition system based on electronic information technology also includes an instruction output module, which is used to output early warning instructions based on the comprehensive risk score and prevention and control plan data; wherein the early warning instructions include a risk level identifier, prevention and control plan data and GPS positioning data; the risk level identifier is a comprehensive risk score mapped to a level I-IV risk identifier.

[0125] The instruction output module is a functional unit that generates and transmits warning instructions based on system analysis results. This module can be implemented using an API or message queue, encapsulating structured data into a transmittable instruction format. Risk level identification discretizes the continuous numerical comprehensive risk score into four-level classification labels. This can be implemented using a segmented threshold mapping algorithm. For example, the score range of 0-25% is set as Level I, 25-50% as Level II, 50-75% as Level III, and 75-100% as Level IV. This classification identification allows for an intuitive expression of risk levels.

[0126] Specifically, after the pest identification module completes pest type and density analysis and the risk assessment module generates a comprehensive risk score, the command output module receives a data packet containing the score results, prevention and control plan, and geographic coordinates. The score results are converted into risk indicators of levels I-IV using preset mapping rules. For example, a score exceeding 75 is marked as a Level IV red alert. Prevention and control plan data is extracted from the policy database to identify biological or chemical control measures that match the pest type, such as an imidacloprid spraying plan for aphids. GPS positioning data is derived from drone flight path records and is spatiotemporally matched with pest occurrence images to generate a set of coordinate points. The final warning output, containing these three types of data, is transmitted to the farmland management terminal via the wireless communication module.

[0127] In some embodiments, risk level identification can be combined with color coding and numerical grading, for example, Level IV corresponds to a red icon and a buzzer alarm. GPS positioning data can be overlaid on electronic maps to display pest distribution heat maps, for example, by color-coding the risk level of each coordinate point on a GIS platform. Control plan data can include a pesticide dosage calculation function, for example, automatically generating a recommended pesticide dosage per mu based on pest population density.

[0128] Through the above technical solution, this application solves the technical problems of dispersed early warning information and disconnection between treatment measures and geographical locations in traditional pest monitoring. Risk level identification enables rapid identification of the degree of harm, combined with GPS positioning data to accurately guide the prevention and control areas, while linking prevention and control plan data to provide treatment measures, forming a complete decision-making support chain. For example, when the system outputs a Level IV warning, the operator can directly locate the northeast corner of the No. 3 field area based on the coordinates and execute the corresponding emergency spraying plan, avoiding response delays caused by information fragmentation.

[0129] like Figure 5 As shown, the present application further proposes that an intelligent crop pest image recognition system based on electronic information technology also includes a user feedback module, which specifically includes:

[0130] Embed image annotation tools for users to correct recognition results;

[0131] Receive user correction data for the identification result; the correction data includes manually marked pest types and location frame coordinates;

[0132] The user feedback module is connected to the closed-loop optimization module, and the correction data is used to update the new data of the closed-loop optimization module.

[0133] The embedded image annotation tool is an interactive annotation component integrated into the system's user interface. It can be implemented using a web-based annotation interface or a mobile canvas tool. It allows users to directly select insect areas and the correct pest species within the original image. This tool uses a coordinate mapping algorithm to convert user input into a standardized data format, resolving the mismatch between manually edited data and machine-recognized results.

[0134] Correction data refers to the correction information provided by the user to the system's automatic recognition results. It may specifically include the insect position coordinates, pest species labels and confidence scores re-labeled by the user. It is transmitted to the back-end database through the JSON data structure to eliminate the accumulation of errors caused by system misjudgments or missed detections.

[0135] The connection between the user feedback module and the closed-loop optimization module refers to the establishment of a data channel between the two modules. Specifically, a RESTful API interface can be used to implement real-time data transmission, so that user correction data can trigger the incremental learning mechanism of the closed-loop optimization module, solving the problem that traditional systems cannot dynamically optimize based on user feedback.

[0136] Specifically, when a user uses an image annotation tool to modify the system's pest identification output, the modified data is encapsulated into a correction data packet containing the image ID, correction timestamp, and annotation content. This data packet is sent to the closed-loop optimization module via an encrypted transmission protocol, triggering an incremental update of the historical pest spatiotemporal database. This new data includes the user's modified pest species label, location coordinates, and corresponding environmental monitoring data, forming a training sample set with verification labels. The closed-loop optimization module uses this new data to fine-tune the parameters of the improved YOLO recognition model. It then uses transfer learning to iteratively train the original model weights, thereby improving the model's recognition accuracy in complex environments.

[0137] Through the above technical solutions, this application can effectively reduce the error rate of prevention and control decisions caused by model misjudgment, and continuously optimize the generalization ability of the recognition model through user-corrected data. At the same time, the closed-loop optimization mechanism associates user feedback data with the historical database, enhances the system's adaptability to different regions, crop varieties, and climate conditions, and forms an intelligent recognition system with the ability of self-evolution.

[0138] The following is a specific example of the intelligent recognition of pest images in a corn planting area in the North China Plain:

[0139] An agricultural cooperative manages 300 mu of corn fields in the North China Plain, mainly planting the Zhengdan 958 variety. In previous years, during the peak period of corn borers and armyworms from July to August, the missed inspection rate of traditional manual inspections was about 30%, and the probability of misjudging armyworms and corn borers depending on the experience of technicians reached 25%; when using a single camera for monitoring, the missed inspection rate of the lower insect bodies due to canopy occlusion exceeded 40%; the early warning only indicated "pest presence" and did not provide targeted prevention and control plans, often resulting in a 10%-15% reduction in production due to delayed prevention and control.

[0140] This system is deployed in this area as follows:

[0141] Image acquisition module: Install 4 groups of fixed multi-spectral camera arrays at the four corners of the corn field (installation height: 3 meters, wide-angle lens coverage radius: 50 meters), and collect visible light RGB images at the first sampling frequency (once every 30 minutes); synchronously deploy 1 multi-rotor drone (preset "field" - shaped flight path, covering the entire area in 40 minutes), and collect multi-angle images at the second sampling frequency (perform 3 full-area scans at 8:00, 14:00, and 18:00 every day), and record GPS positioning data (accuracy ±0.5 meters), environmental monitoring data (temperature 28 ± 2°C, humidity 75 ± 5%, light intensity 8000 ± 500 lux), and shooting time (accurate to seconds).

[0142] At 14:00 on July 15, 2024, the drone completed the second full-area scan and obtained the RGB image (resolution 4032×3024) of the southwest corner of the corn field (GPS coordinates: N37°25′12″, E114°38′45″) and the associated environmental data (temperature 29°C, humidity 78%, light intensity 8200 lux). The preprocessing module performs the following operations:

[0143] Image normalization: Scale the original image to 640×640 pixels through the bilinear interpolation algorithm, keeping the leaf edges and insect body contours clear;

[0144] Use the CLAHE algorithm to enhance the local contrast of the image, eliminating the local dark areas caused by corn canopy occlusion (the average gray value of the original dark area is increased from 35 to 60, and the standard deviation is decreased from 15 to 8);

[0145] Applying the non-local means denoising method, the image noise standard deviation was reduced from 12 to 5 while preserving the insect body texture (such as the corn borer body segment stripes);

[0146] By matching the timestamps (the drone shooting time is 14:02:35 and the environmental sensor recording time is 14:02:36), the processed image matrix is ​​bound with the environmental data and GPS coordinates into a standardized data packet (tensor dimension: [1, 640,640, 3]).

[0147] The standardized data packet is input into the improved YOLO recognition model, which performs the following operations:

[0148] The environmental monitoring data (temperature 29°C, humidity 78%, light 8200 lux) is mapped to a 640×640×3 weight matrix (temperature weight 0.4, humidity 0.3, light 0.3) through a 1×1 convolution kernel. This is then multiplied element-by-element with the image matrix to enhance the characteristic response of corn borer larvae (which prefer humidity) in a high humidity environment.

[0149] The pest time series sample library was called and the historical features of the current image (the body widths of corn borer larvae were 2 mm, 3 mm, and 4 mm, respectively) were compared with the previous three frames (13:32, 14:02, and 14:32) using the cosine similarity algorithm. The confidence level of the evolutionary stage was calculated (0.85, indicating an "outbreak period").

[0150] The image quality (0.82) was calculated by weighting the image clarity score (edge ​​gradient mean 0.75) and the illumination uniformity score (regional grayscale variance 0.12), which was higher than the dynamic threshold (0.78, triggered by an illumination intensity of 8200 lux), without the need for re-sampling data.

[0151] The model outputs the pest type as corn borer (confidence level 0.92), and 12 coordinates are selected for the insect's location (distributed in the middle and lower part of the corn stalk).

[0152] Use the environmental suitability comparison table to find the optimal environmental range for corn borers (temperature 25-30°C, humidity 70-80%, light 7000-9000 lux) and calculate the individual scores:

[0153] Temperature score: |29-27.5| / (5 / 2)=0.6 → 1-0.6=0.4 (median 27.5°C, range 5°C);

[0154] Humidity score: |78-75| / (10 / 2)=0.6 → 1-0.6=0.4 (median 75%, range 10%)

[0155] Light score: |8200-8000| / (2000 / 2)=0.2 → 1-0.2=0.8 (median 8000 lux, range width 2000 lux); weighted sum (weights 0.4, 0.3, 0.3) to get the environmental suitability .

[0156] The insect population density was calculated (12 selected coordinates correspond to an area of ​​0.8 m2, density = 12 / 0.8 = 15 insects / m2), and the risk coefficient (basic risk coefficient for corn borer is 0.6), environmental suitability (0.52), and pest development status (outbreak score 0.7) was integrated to calculate the following formula: (The preset threshold is 1.0), triggering an early warning.

[0157] Because the comprehensive scores of the previous two scans (13:32 and 14:02) in the same area (southwest corner) were 1.05 and 1.12 respectively (both exceeding the threshold), an upgraded prevention and control plan was generated:

[0158] Environmental suitability S=0.52<0.8, chemical control (spraying chlorantraniliprole) is preferred;

[0159] If the insect population density is 15 insects / ㎡> the density threshold (10 insects / ㎡), the concentration of the pesticide will be automatically adjusted from 1000 times to 800 times;

[0160] After searching the control strategy database, the historical effectiveness of the same plan (insect population reduction rate within 7 days) was 75%>60%, so this plan was directly adopted.

[0161] The command output module generates an early warning command: the risk level is marked as Level III (a score of 1.24 corresponds to the 75-100% range), highlighted in red; the prevention and control plan is to spray the southwest corner area (N37°25′12″, E114°38′45″) with 800 times diluted chlorantraniliprole, with a dosage of 50ml per mu; the GPS positioning marks the area as a red hot spot on the electronic map, and displays the risk trend in the past 7 days (the score has increased from 1.24 to 1.24). ).

[0162] After spraying, the user (plant protection worker) corrected the recognition results through the user feedback module: 11 of the 12 insects were confirmed to be corn borers (1 was an insecticide), and the coordinates of one frame selection were manually adjusted (offset 2 cm from the original position). The closed-loop optimization module incrementally updated the corrected data (image ID: 20240715_1402, actual pest types: corn borer (11), insecticide (1), corrected coordinates) to the historical pest spatiotemporal database, and adjusted the corn borer risk coefficient (from 0.6 to 0.65) and the insecticide risk coefficient (newly added to 0.5). At the same time, the new data was used to fine-tune the parameters of the improved YOLO model (freezing the Backbone layer and training the Neck layer). The model's recognition accuracy for corn borers increased from 92% to 95%, and the recognition accuracy for insecticides increased from 0 (originally untrained) to 85%.

[0163] This system dynamically triggers updated control plans based on comprehensive scoring. Within three days of spraying, the insect population dropped from 15 to 3 insects / ㎡ (an 80% decrease), preventing the spread of insect swarms during an outbreak (where the population could reach over 30 insects / ㎡). After the closed-loop optimization module updated model parameters based on user feedback, the missed detection rate for corn borers dropped from 8% to 3% over the following two weeks. The system also improved its ability to identify armyworms from scratch (with an accuracy rate of 85%), adapting to the emergence of new pest types in the field.

[0164] This example fully demonstrates the application process of the system in cornfield pest monitoring. Through multi-source data fusion, dynamic risk assessment and closed-loop optimization, it effectively solves the problems of data isolation, high misjudgment rate and delayed decision-making in traditional methods, and realizes precise pest identification, forward-looking early warning and adaptive evolution of the system.

[0165] The technical scope of the present invention is not limited to the contents of the above description. Those skilled in the art can make various deformations and modifications to the above embodiments without departing from the technical idea of ​​the present invention, and these deformations and modifications should all fall within the protection scope of the present invention.

Claims

1. An intelligent crop pest image recognition system based on electronic information technology, characterized by: include: An image acquisition module is used to acquire RGB image data of the target crop area and simultaneously acquire environmental monitoring data and image capture time data associated with the RGB image data; A preprocessing module, configured to preprocess the RGB image data, the environmental monitoring data, and the image capture time data to generate a processed image matrix, environmental monitoring processing data, and image capture time processing data; An insect pest recognition module is used to fuse the image matrix with the environmental monitoring data through an improved YOLO recognition model, and output insect pest type data and insect body location frame selection coordinates; a comparison module for comparing the image shooting time processing data with pest evolution stage data in a pre-stored historical pest spatiotemporal database to generate pest development status data; A risk assessment module is used to call a preset pest hazard level dictionary to obtain a risk coefficient matching the pest type data, and calculate an environmental suitability value in combination with the environmental monitoring and processing data; A risk scoring module is used to collect insect population density data based on the insect body position frame coordinates, and to generate a comprehensive risk score by integrating the insect pest development status data, the risk coefficient, the environmental suitability value, and the insect population density data; The early warning decision module is used to call the pre-stored prevention and control strategy database to match the corresponding prevention and control plan data according to the pest type data when the comprehensive risk score exceeds the preset risk threshold.

2. The intelligent crop pest image recognition system based on electronic information technology according to claim 1, characterized in that: The image acquisition module includes: A fixed multispectral camera array is deployed at a preset height in the farmland to collect the RGB image data in the visible light band at a first sampling frequency; The drone is called to capture multi-angle images at a second sampling frequency along a preset flight path to obtain GPS positioning data associated with the RGB image data, the environmental monitoring data, and the image shooting time data; wherein the environmental monitoring data includes temperature, humidity, and light intensity.

3. The intelligent crop pest image recognition system based on electronic information technology according to claim 2, characterized in that: The RGB image data, the environmental monitoring data, and the image shooting time data are preprocessed to generate a processed image matrix, environmental monitoring processing data, and image shooting time processing data. The specific steps of the preprocessing include: Scaling the input RGB image data to 640×640 pixels uniformly using a bilinear interpolation algorithm; Performing illumination compensation on the RGB image data using the CLAHE algorithm; Applying a non-local means denoising method to eliminate image noise of the RGB image data; Aligning the processed RGB image data with the GPS positioning data, the environmental monitoring data, and the image capture time data through a timestamp; Output the processed image matrix, GPS positioning processing data, environmental monitoring processing data and image shooting time processing data; the processed data is a tensor data structure that meets the input requirements of the improved YOLO recognition model.

4. The crop pest image intelligent recognition system based on electronic information technology according to claim 1, characterized in that: The specific steps of fusing the image matrix with the environmental monitoring data through the improved YOLO recognition model to output the pest species data and the insect body location frame selection coordinates include: Mapping the environmental monitoring processed data into a weight matrix of the same dimension as the image matrix through a 1×1 convolution kernel, and multiplying it element-by-element with the RGB channels of the image matrix; Associate the current image matrix with the pre-stored pest time series sample library, use the cosine similarity algorithm to compare the historical pest features of the previous three frames and perform confidence matching, and output the confidence of the pest evolution stage; The image matrix is ​​scored for clarity and illumination uniformity to obtain image quality. When the image quality is lower than a quality threshold, the image acquisition module is triggered to reshoot, and the quality threshold is dynamically adjusted by weighted calculation of the image quality.

5. The crop pest image intelligent recognition system based on electronic information technology according to claim 1, characterized in that: The preset pest hazard level dictionary is called to match the risk coefficient of the pest type data, and the environmental suitability value is calculated in combination with the current environmental monitoring data. The calculation method of the environmental suitability value is: Query the optimal range of the target pest corresponding to the pest type data from the preset environmental suitability comparison table , , ; The suitability score of a single environmental factor is calculated using a piecewise function. The formula is as follows: ； in, is the actual monitoring value of the environmental factor, is the median value of the optimal range, is the optimal range width; The weighted summation generates the environmental suitability value: ； in, is the environmental suitability value, 、 、 They are the temperature factor single score, humidity factor single score, light factor single score, and environmental factor weight Obtained through training on historical pest outbreak data.

6. The crop pest image intelligent recognition system based on electronic information technology according to claim 1, characterized in that: The insect population density data is collected based on the insect body position frame coordinates, and the pest development status data, the risk coefficient, the environmental suitability value, and the insect population density data are integrated to generate a comprehensive risk score. The generation of the comprehensive risk score specifically includes: Establish a preset pest hazard level dictionary to store the basic risk factors of various pests in the form of key-value pairs ; According to the pest type data output by the pest identification module, the pest hazard level dictionary is searched and the basic risk coefficient is used to Get the corresponding risk factor ; Based on the insect body position, select the coordinates and count the number of insects per unit area to obtain the insect population density data. ; The final comprehensive risk score fusion formula is calculated as follows: ； in is the preset weight coefficient, satisfying ; is the environmental suitability value; Develop status data for the pest.

7. The intelligent crop pest image recognition system based on electronic information technology according to claim 6, characterized in that: When the comprehensive risk score exceeds a preset risk threshold, the pre-stored control strategy database is called according to the pest type data to match the corresponding control plan data. The control plan data matching process further includes: When it is detected that the comprehensive risk score in the same area exceeds the preset risk threshold for three consecutive times, an upgraded prevention and control plan is generated; The upgraded prevention and control schemes are prioritized, and the rules output a combination of prevention and control schemes with validity periods: (1) Current environmental suitability values When >0.8, biological control is preferred; (2) After the biological control is initiated, if the insect population density data Density continues to increase, chemical control is added; (3) The insect population density data of the current number of insects per unit area Automatically increase the chemical concentration ratio when the density is greater than the threshold; (4) When the historical efficiency of the same solution is less than 60%, start searching for alternative solutions.

8. The intelligent crop pest image recognition system based on electronic information technology according to claim 7, characterized in that: The crop pest image intelligent recognition system based on electronic information technology further includes a closed-loop optimization module, which performs the following operations: Obtain confirmation data and false alarm marks submitted by the user terminal, where the confirmation data includes the image ID, the actual pest type, and the corrected insect body location frame coordinates; Based on the confirmed data, data feedback is performed to obtain new data, and the historical pest spatiotemporal database is incrementally updated to supplement new pest samples with spatiotemporal labels; Update the pest hazard level dictionary and adjust the risk coefficient according to the actual hazard record; update the control strategy database and associate the records with the effectiveness of executing the control plan data; the effectiveness is the insect population reduction rate after the same plan is executed within 7 days extracted from the control strategy database; The newly added data is called to fine-tune the parameters of the improved YOLO recognition model, and the model parameters of the improved YOLO recognition model are updated.

9. The crop pest image intelligent recognition system based on electronic information technology according to claim 2, characterized in that: The intelligent crop pest image recognition system based on electronic information technology also includes an instruction output module, which is used to output an early warning instruction based on the comprehensive risk score and the prevention and control plan data; wherein the early warning instruction includes a risk level identifier, the prevention and control plan data and the GPS positioning data; the risk level identifier is the comprehensive risk score mapped to a level I-IV risk identifier.

10. The crop pest image intelligent recognition system based on electronic information technology according to claim 8, characterized in that: The crop pest image intelligent recognition system based on electronic information technology further includes a user feedback module, which specifically includes: By embedding an image annotation tool for users to correct the recognition results; Receive correction data of the identification result from the user; wherein the correction data includes manually marked pest types and location frame coordinates; The user feedback module is connected to the closed-loop optimization module, and the correction data is used to update the newly added data of the closed-loop optimization module.

Images