Pest and disease early warning method and system based on plant monitoring

Abstract

The invention discloses a plant disease and insect pest early warning method and system based on plant monitoring, and belongs to the field of plant disease and insect pest early warning, and the method comprises the steps: extracting the contour of a lesion region through an edge detection algorithm, carrying out the smoothing of the contour through the combination of morphological operation, and obtaining a more precise lesion region boundary; according to an image segmentation result and a disease type identification result, a plant health condition evaluation model is established, and the plant damage degree is quantified; environmental data and image data are acquired from a plurality of sensor nodes distributed in a field, and the data are gathered to a regional gateway through wireless transmission; carrying out preprocessing and feature extraction on the converged heterogeneous data in a regional gateway, removing noise data and redundant information, and extracting key features; and carrying out pest detection and counting on the preprocessed image by using a deep learning model, carrying out modeling on a pest number change trend, predicting population density change in a period of time in the future in combination with environmental factors, and generating a detection result.

Description

Technical Field

[0001] The present invention belongs to the field of plant disease and insect pest early warning, and in particular relates to a disease and insect pest early warning method and system based on plant monitoring. Background Art

[0002] Plant pest and disease early warning systems play a vital role in modern agricultural production and are of great significance for ensuring food security and sustainable agricultural development. However, most current early warning systems still suffer from insufficient monitoring accuracy, inefficient data integration, and poor warning timeliness, making them unable to meet the needs of precision agriculture. In the field of plant pest and disease monitoring, accurate identification and quantification of pest and disease conditions are the foundation for effective early warning. Traditional methods rely primarily on manual observation and simple image analysis, which are unable to capture subtle changes in pests and diseases and their rapid spread.

[0003] At the same time, the dispersed nature and diverse formats of monitoring data make it difficult to efficiently aggregate and analyze multi-source, heterogeneous data, hindering the overall performance of the early warning system. Faced with these challenges, improving the accuracy of pest and disease image segmentation, efficiently aggregating multi-source, heterogeneous data, and accurately quantifying pest and disease populations have become key issues that require in-depth research. Particularly in complex field environments, achieving accurate identification and quantitative analysis of pests and diseases for different crops and at different growth stages, as well as rationally allocating computing resources between the edge and the cloud to improve the system's real-time performance and scalability, are key technical challenges that require in-depth research. Summary of the Invention

[0004] To solve the above technical problems, the present invention provides a plant pest and disease early warning method based on plant monitoring, comprising:

[0005] A convolutional neural network is used to perform pixel-level segmentation on plant leaf images, dividing the images into healthy tissue areas and diseased areas to obtain image segmentation results.

[0006] The contour of the lesion area is extracted based on the edge detection algorithm, and the contour is smoothed by combining morphological operations to obtain the boundary of the lesion area;

[0007] Use color and texture features to perform cluster analysis on the segmented lesion area to distinguish different types of disease symptoms and obtain disease type recognition results;

[0008] Establishing a plant health assessment model based on the image segmentation results and the disease type identification results to quantify the degree of plant damage;

[0009] Acquire environmental data and image data from multiple sensor nodes distributed in the field, and aggregate the data to the regional gateway through wireless transmission;

[0010] At the regional gateway, the aggregated heterogeneous data is preprocessed and feature extracted to remove noise data and redundant information to obtain preprocessed images.

[0011] Using a deep learning model to detect and count pests in the preprocessed images, identify pest individuals at different developmental stages, and count the number of each type of pest;

[0012] Based on the time series analysis method, the pest population change trend is modeled and the pest population density change in the future is predicted in combination with environmental factors;

[0013] A pest and disease risk assessment model is constructed based on the plant damage degree and the insect population density change, and pest and disease detection is performed based on the pest and disease risk assessment model to obtain detection results.

[0014] Preferably, the process of obtaining the image segmentation result includes:

[0015] Perform pixel-level segmentation on plant leaf images through convolutional neural networks to obtain preliminary segmentation areas;

[0016] Based on the preliminary segmented area, the healthy tissue area and the diseased area are divided by threshold judgment, and the boundaries of the two areas are determined;

[0017] Obtain the number of pixels from the healthy tissue area and the lesion area, and calculate the total pixel value of each area;

[0018] The pixel value of the total leaf area is extracted from the plant leaf image to obtain the overall area data;

[0019] The image segmentation result is obtained based on the overall area data.

[0020] Preferably, the process of obtaining the boundary of the lesion area includes:

[0021] Obtaining an initial contour of the lesion area from the image segmentation result by using an edge detection algorithm to obtain first contour data;

[0022] Using morphological operations to smooth the first contour data to obtain second contour data;

[0023] For the second contour data, if the boundary noise is detected to exceed the preset threshold, the Gaussian filtering algorithm is used for optimization to obtain the third contour data;

[0024] According to the third contour data, a coordinate set of boundary points is obtained to determine the precise boundary of the lesion area;

[0025] The precise boundary of the lesion area is verified by the regional segmentation algorithm to obtain the segmented regional data;

[0026] For the segmented area data, the contour tracking method is used to extract the final boundary and obtain the optimized contour data;

[0027] Based on the optimized contour data, the boundary integrity is judged through geometric calculation to obtain the final lesion area boundary.

[0028] Preferably, the process of obtaining the disease type identification result includes:

[0029] Processing the boundary of the lesion area using a color feature extraction method to obtain color feature data;

[0030] Using a texture feature extraction method to analyze the boundary of the lesion area to obtain texture feature data;

[0031] The color feature data and texture feature data are clustered and analyzed using the K-means clustering algorithm to obtain the symptom classification results;

[0032] The disease type identification result is obtained based on the symptom classification result.

[0033] Preferably, the process of counting the number of various types of pests includes:

[0034] Initial data is obtained by preprocessing the image, and a deep learning model is used to extract features from the image to obtain preliminary pest positioning;

[0035] Based on the preliminary positioning results, the detection model is used to analyze each area to determine the distribution of individual pests;

[0036] If individual pests are detected, features are extracted based on image analysis to determine the developmental stage classification;

[0037] Through the stage classification results, the pest individuals at each developmental stage are obtained and the corresponding number is counted;

[0038] The quantitative statistical method is used to count the individuals of each category of pests and obtain category statistical data;

[0039] Extract distribution patterns from category statistics to determine complete results of pest detection;

[0040] Based on the complete results, data processing techniques are used to generate structured output and obtain final statistical information.

[0041] Preferably, the process of predicting insect population density changes in the future period of time in combination with environmental factors includes:

[0042] Obtain historical pest population data and corresponding environmental factor data, construct a time series data set, preprocess the time series data, including data cleaning, normalization and stationarity test; use the autoregressive moving average model to model the pest population time series, and obtain preliminary prediction results; establish a multiple regression model based on the environmental factor data to analyze the impact of environmental factors on the number of pests; construct an integrated prediction model by combining the time series model and the multiple regression model; optimize the parameters of the integrated prediction model, minimize the prediction error, and improve the model accuracy; use the optimized integrated model to predict the changes in insect population density in the future.

[0043] Preferably, the process of obtaining the test result includes:

[0044] The plant damage degree and pest density data are acquired through sensors to determine the initial state parameters;

[0045] Regression analysis was used to process the data on the degree of damage and density to obtain the characteristics of the changing trend;

[0046] Build a risk assessment model based on the characteristics of the changing trend to obtain the risk distribution of pests and diseases;

[0047] Conduct pest and disease detection based on risk distribution and identify high-risk areas;

[0048] If the high-risk area exceeds the preset threshold, the damage level image is analyzed through a convolutional neural network to determine the specific type of pest and disease;

[0049] Obtain test results and update risk assessment model parameters based on change trends;

[0050] The updated model is used to detect a new round of pests and diseases and obtain optimized detection results.

[0051] On the other hand, the present invention also provides a plant pest and disease early warning system based on plant monitoring, comprising:

[0052] An image segmentation module is used to perform pixel-level segmentation of plant leaf images using a convolutional neural network, dividing the image into healthy tissue areas and diseased areas, and calculating the proportion of the diseased area to the total leaf area based on the segmentation results;

[0053] The contour processing module is used to extract the contour of the lesion area through the edge detection algorithm and smooth the contour in combination with morphological operations to obtain a more accurate boundary of the lesion area;

[0054] Cluster analysis module, used to perform cluster analysis on the segmented lesion area using color features and texture features;

[0055] The health assessment module is used to establish a plant health assessment model based on image segmentation results and disease type identification results to quantify the degree of plant damage;

[0056] The data acquisition module is used to obtain environmental data and image data from multiple sensor nodes distributed in the field, and aggregate the data to the regional gateway through wireless transmission;

[0057] The data preprocessing module is used to preprocess and extract features from the aggregated heterogeneous data at the regional gateway, remove noise data and redundant information, and extract key features;

[0058] The pest detection module uses a deep learning model to detect and count pests in preprocessed images, identify pests at different developmental stages, and count the number of each type of pest.

[0059] The trend prediction module is used to model the pest population change trend based on time series analysis methods and predict the pest population density changes in the future by combining environmental factors;

[0060] The risk assessment module is used to comprehensively analyze the degree of plant damage, pest density and its changing trends, build a pest and disease risk assessment model, perform pest and disease detection based on the pest and disease risk assessment model, and obtain detection results.

[0061] On the other hand, the present invention also provides an electronic device, comprising a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the method is implemented when the processor executes the computing program.

[0062] On the other hand, the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program implements the method when executed by a processor.

[0063] Compared with the prior art, the present invention has the following advantages and technical effects:

[0064] The present invention discloses an intelligent plant pest and disease monitoring method based on image analysis and sensor networks. The method uses a convolutional neural network to perform pixel-level segmentation on plant leaf images, dividing the images into healthy and diseased areas, and extracting precise lesion contours through edge detection and morphological processing. Combining color and texture features, cluster analysis is performed on the diseased areas to identify different types of disease symptoms. At the same time, the present invention uses sensor nodes distributed in the field to collect environmental and image data, and performs data preprocessing and feature extraction at the regional gateway. Pests are detected and counted in the images using a deep learning model, and changes in insect population density are predicted based on time series analysis. Finally, a comprehensive analysis of the degree of plant damage, pest density, and its changing trends is performed to construct a pest and disease risk assessment model to achieve intelligent pest and disease detection and early warning. The present invention can effectively improve the accuracy and efficiency of plant pest and disease monitoring and provide strong support for agricultural production management. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0066] Figure 1 Schematic diagram of a method flow in an embodiment of the present invention. DETAILED DESCRIPTION

[0067] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0068] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0069] Example 1

[0070] like Figure 1 As shown, this embodiment provides a plant pest and disease early warning method based on plant monitoring, including:

[0071] S101. Use a convolutional neural network to perform pixel-level segmentation on the plant leaf image, divide the image into healthy tissue area and diseased area, and calculate the proportion of the diseased area to the total leaf area based on the segmentation results.

[0072] A convolutional neural network is used to perform pixel-level segmentation on the plant leaf image to obtain preliminary segmented regions. Based on the preliminary segmented regions, a threshold is used to demarcate healthy tissue regions and diseased regions, and the boundaries between the two regions are determined. The number of pixels in the healthy tissue region and the diseased region is obtained, and the total pixel value of each region is calculated. The pixel value of the total leaf area is extracted from the plant leaf image to obtain the overall area data. The lesion ratio value is calculated by dividing the pixel value of the lesion region by the pixel value of the total leaf area. If the lesion ratio value exceeds the preset threshold, the convolutional neural network is used to optimize the boundary of the lesion region to obtain a precise segmentation result. The pixel value of the lesion region is updated based on the precise segmentation result, and the final lesion ratio is determined by recalculation.

[0073] Specifically, a convolutional neural network (UNet) was first used to perform pixel-level segmentation on plant leaf images. The UNet segmentation model was selected, with an input image resolution of 512×512 pixels. The model consisted of four downsampling layers and four upsampling layers, each with a convolution kernel size of 3×3 and a stride of 1. The model was trained using the Adam optimizer with a learning rate of 0.01, a batch size of 8, and 100 iterations. The training set consisted of 1,000 annotated leaf images, with healthy tissue labeled as 1 and lesioned areas as 0. During training, the model parameters were optimized using a backpropagation algorithm using a cross-entropy loss function. After segmentation, the output images were post-processed using morphological operations to remove noise and retain connected components with an area greater than 50 pixels. Next, the number of pixels in the lesioned area was counted. Assuming the segmentation result indicated a lesioned area of 5,000 pixels and a total leaf area of 20,000 pixels, the proportion of the lesioned area to the total leaf area was 25%. To verify the segmentation performance, the DICE coefficient was used as an evaluation metric, with a value of 85, indicating high segmentation accuracy. Finally, the segmentation results and proportional data are stored in the database for subsequent analysis and decision support.

[0074] S102. Extract the contour of the lesion area using an edge detection algorithm, and smooth the contour using morphological operations to obtain a more accurate boundary of the lesion area.

[0075] The initial contour of the lesion area is obtained from the original image through the edge detection algorithm to obtain the first contour data. The first contour data is smoothed by morphological operations to obtain the second contour data. For the second contour data, if the boundary noise is detected to exceed the preset threshold, it is optimized by the Gaussian filtering algorithm to obtain the third contour data. Based on the third contour data, the coordinate set of the boundary points is obtained to determine the precise boundary of the lesion area. The precise boundary of the lesion area is verified by the region segmentation algorithm to obtain the segmented region data. For the segmented region data, the final boundary is extracted by the contour tracking method to obtain the optimized contour data. Based on the optimized contour data, the boundary integrity is judged by geometric calculation to obtain the final lesion area contour.

[0076] Specifically, in image processing, the edge detection algorithm can extract the contour of the lesion area through the Canny operator. First, the input image is Gaussian filtered with a standard deviation of 5 to smooth the image and reduce noise. Next, the Sobel operator is used to calculate the gradient amplitude and direction of the image, with thresholds set to 50 and 150 to determine strong and weak edges. The local maximum of the edge is retained and pseudo-edges are removed through the non-maximum suppression method. Finally, the preliminary contour of the plant lesion area is obtained through double threshold detection and edge connection. In order to improve the accuracy of the contour, morphological operations can be used for smoothing. A 3×3 square structuring element is used for closing operations to fill small holes in the contour and smooth the edges. Then, an opening operation is applied to remove small isolated points and burrs, and the structuring element is also 3×3. Through morphological reconstruction, the main boundaries of the lesion area are retained while irrelevant noise is removed. To further optimize the boundary, an algorithm based on the active contour model, such as the Snake model, can be used. The initial contour is used as input, the contour point position is iteratively updated, the elastic coefficient in the energy function is set to 1, the rigidity coefficient is set to 2, the external force coefficient is set to 7, and the number of iterations is 100 to obtain a smoother and more accurate boundary of the plant lesion area.

[0077] S103. Perform cluster analysis on the segmented diseased areas using color features and texture features to distinguish different types of disease symptoms, such as chlorosis, necrosis, and spots.

[0078] Image processing techniques are used to segment the original image and obtain the lesion area. The lesion area is processed using color feature extraction methods to obtain color feature data. Texture feature extraction methods are used to analyze the lesion area to obtain texture feature data. K-means clustering is used to perform cluster analysis on the color and texture feature data to obtain symptom classification results. If the green component in the color feature data is below a preset threshold, it is judged as a chlorotic symptom and the chlorotic symptom area is determined. If the regularity in the texture feature data is below a preset threshold, it is judged as a necrotic symptom and the necrotic symptom area is determined. Based on the cluster analysis results and judgment criteria, the spot symptom area is obtained to distinguish the disease type.

[0079] Specifically, based on the segmentation results of the diseased area, color and texture features are first extracted to distinguish different types of disease symptoms. For color features, the HSV color space is used to extract the mean, variance, and histogram distribution of hue (H), saturation (S), and brightness (V).

[0080] For example, the mean hue of chlorotic areas typically ranges from 50 to 70, while the mean hue of necrotic areas may be below 30. Texture features are calculated using the gray-level co-occurrence matrix (GLCM) to measure contrast, correlation, and energy. For example, the contrast value of spots may be higher than 2.5, while the contrast value of chlorotic areas is typically lower than 1.0. Next, these features are clustered using the K-means clustering algorithm, with the number of clusters set to 3 to distinguish between chlorosis, necrosis, and spots. The similarity between feature vectors is calculated using the Euclidean distance, and the cluster centers are iteratively optimized until convergence. After clustering, statistical analysis is performed on each cluster. For example, the mean hue of the chlorotic cluster is 60 with a variance of 10, while the mean hue of the necrotic cluster is 25 with a variance of 5, further verifying the accuracy of the clustering results. Finally, a disease distribution map is generated based on the clustering results, providing data support for disease diagnosis and prevention.

[0081] S104: Based on the image segmentation results and the disease type identification results, a plant health assessment model is established to quantify the degree of plant damage.

[0082] Image segmentation technology is used to obtain segmented data, separating plant regions from the original image. Image analysis is performed on the segmented data to extract characteristic information about the plant regions. Based on this characteristic information, the disease type is identified and a classification result is determined. If the classification result indicates the presence of a disease, a pre-set threshold is used to determine the damage index. After obtaining the damage index, an assessment model is constructed to calculate a quantitative plant health value. A linear regression algorithm is used to adjust the quantization level of the health value to obtain a final damage level. This final damage level is then used to determine the plant health classification result.

[0083] Specifically, in the image segmentation stage, a U-Net model from deep learning was used to segment plant leaf images. The model input was a 256×256 pixel RGB image, and the output was a binary segmentation map, where a pixel value of 1 represented the leaf area and 0 represented the background area. The model was optimized using a cross-entropy loss function, achieving a Dice coefficient of 92 on the test set. In the disease identification stage, a ResNet-50 network was used to classify the segmented leaf areas. The input was a 224×224 pixel cropped image, and the output was a probability distribution of five categories: healthy, brown spot, and powdery mildew. The model used the Adam optimizer with a learning rate set to 0.01, achieving an accuracy of 89% on the validation set. In the health assessment stage, the proportion of leaf damaged area was calculated based on the segmentation results, and different weights were assigned based on the disease type.

[0084] For example, areas identified as brown spot have a weight of 8, areas identified as powdery mildew have a weight of 6, and healthy areas have a weight of 0. The overall damage score is obtained through weighted summation using the formula: score = Σ(area × weight) / total leaf area × 100. If a leaf has 10% brown spot and 15% powdery mildew, its damage score is (1 × 8 + 15 × 6) × 100 = 17. Finally, plant health is categorized into five levels: excellent, good, moderate, poor, and critical, based on the score intervals [0, 20], [20, 40], [40, 60], [60, 80], and [80, 100], enabling a quantitative assessment of plant damage.

[0085] S105 , acquiring environmental data and image data from multiple sensor nodes distributed in the field, and aggregating the data to a regional gateway via wireless transmission.

[0086] Sensor nodes collect field environmental parameters and image data, and pre-process and preliminarily analyze the collected environmental data. Sensor nodes send data to the gateway through wireless communication modules. The gateway receives data sent by multiple sensor nodes and summarizes it. The gateway classifies and organizes the summarized data and converts the format. It uses data compression algorithms to compress the organized data. The gateway sends the compressed data packets to the data center through the wide area network.

[0087] Specifically, multiple sensor nodes are distributed throughout the fields, each equipped with various sensors for temperature, humidity, light intensity, and other data. They collect environmental data every five minutes and are equipped with cameras to capture image data every ten minutes. The sensor nodes utilize LoRa wireless communication technology, with a spreading factor of SF=7, and transmit data in the 433MHz frequency band, with a transmission range of up to three kilometers. Each packet is 51 bytes in size, and a CRC-16 checksum is used to ensure data integrity. A regional gateway is deployed in the center of the field and receives data uploaded by each node using the MQTT protocol. The gateway has a built-in edge computing module and uses the K-means clustering algorithm to analyze temperature data in real time. It issues an immediate warning if a temperature anomaly exceeding 3 degrees Celsius is detected in a specific area. Image data is compressed using the JPEG algorithm to keep each image size under 200KB. The SSIM algorithm is used to analyze the similarity of consecutive images. When the similarity falls below 0.8, the abnormal image acquisition mechanism is triggered, increasing the sampling frequency to once per minute. After all data is pre-processed at the gateway, it is uploaded to the cloud server via the 4G network, permanently stored using the Hadoop distributed storage system, and trend analysis of historical data is performed using the Spark stream processing framework.

[0088] S106. Preprocess and extract features from the aggregated heterogeneous data at the regional gateway to remove noise data and redundant information and extract key features.

[0089] The receiving gateway acquires multi-source heterogeneous data streams and uses a data classification algorithm to perform preliminary classification and labeling of the data. Preprocessing methods are selected based on the data type, normalizing numerical data and performing word segmentation on text data. A filtering algorithm is used to filter high-frequency noise from the data, resulting in a smoothed data sequence. Correlation analysis is used to calculate the correlation coefficient between data features. If the correlation coefficient exceeds a threshold, the feature is identified as redundant and deleted. Principal component analysis is used to reduce the dimensionality of the preprocessed data and extract the main feature vectors as key features. The extracted key features are encoded and compressed to generate feature vectors. A data fingerprint is constructed based on the compressed feature vectors for subsequent rapid retrieval and matching.

[0090] Specifically, during the preprocessing and feature extraction process of the aggregated heterogeneous data at the regional gateway, a sliding window algorithm is first used to segment the data, with a window size of 100 data points and a step size of 50 data points to ensure data continuity and integrity. Subsequently, a wavelet transform is used to denoise the data, selecting the Daubechies wavelet basis function, decomposing it to four layers, and setting a threshold of 0.5 to remove high-frequency noise while retaining useful information. Next, principal component analysis (PCA) is applied to reduce the data dimension, retaining the first three principal components with an explained variance of 95%, effectively removing redundant information. In the feature extraction stage, a fast Fourier transform (FFT) is used to convert the time-domain data into the frequency domain, extracting key features with a frequency range of 1 Hz to 10 Hz. These features are then classified using a support vector machine (SVM) algorithm, using the radial basis function (RBF) kernel with parameters C set to 0 and gamma set to 0.1 to achieve high-precision feature recognition. This series of processes ultimately extracts key features that not only reduces the data volume but also improves the efficiency and accuracy of subsequent analysis.

[0091] S107. Use a deep learning model to detect and count pests in the preprocessed images, identify pest individuals at different developmental stages, and count the number of each type of pest.

[0092] Initial data is obtained by preprocessing the image. A deep learning model is then used to extract features from the image to obtain preliminary pest location. Based on this preliminary location, the detection model analyzes each area to determine the distribution of individual pests. If individual pests are detected, features are extracted based on image analysis to determine their developmental stage classification. Based on the stage classification results, pests in each developmental stage are identified and counted. Using quantitative statistics, pests of each category are counted to obtain category-specific statistics. Distribution patterns are extracted from the category-specific statistics to determine the complete pest detection results. Based on this complete result, data processing techniques are used to generate a structured output to obtain final statistical information.

[0093] Specifically, during the preprocessing phase, the input image is first normalized and resized to 512×512 pixels. Gaussian filtering with a kernel size of 3×3 is then used to remove noise. Next, histogram equalization is used to enhance image contrast, making pest features more distinct. The YOLOv5 algorithm is used for pest detection. The model is pretrained on the COCO dataset and fine-tuned on a custom dataset of 5,000 annotated images. Training is performed with a learning rate of 0.01, a batch size of 16, and 100 iterations. After model training, detection results are filtered using non-maximum suppression (NMS), with a confidence threshold of 5 and an IoU threshold of 45 to ensure detection accuracy. During the pest recognition phase, the model can distinguish between pests at different developmental stages, such as larvae, pupae, and adults. A classifier outputs the class probability for each detection box and counts the number of each pest class. For pest counting, the system uses overlapping bounding box detection to avoid duplicate counting. Image segmentation technology is also used to finely segment dense areas, achieving over 95% accuracy. Ultimately, the system outputs the specific number of each pest type in each image and generates statistical reports, providing data support for subsequent pest control efforts.

[0094] S108. Based on the time series analysis method, the pest population change trend is modeled and the changes in pest population density in the future are predicted in combination with environmental factors.

[0095] Obtain historical pest population data and corresponding environmental factor data, construct a time series data set, preprocess the time series data, including data cleaning, normalization and stationarity test, use the autoregressive moving average model to model the pest population time series, and obtain preliminary prediction results. Based on the environmental factor data, establish a multiple regression model to analyze the impact of environmental factors on the pest population. By combining the time series model and the multiple regression model, construct an integrated prediction model. Optimize the parameters of the integrated prediction model to minimize the prediction error and improve the model accuracy. Use the optimized integrated model to predict the changes in insect population density in the future.

[0096] Specifically, first, collect historical data on pest populations and environmental factors, such as monthly pest populations, temperature, humidity, and rainfall over the past five years. Assume the pest population data is [100, 150, 200, 250, 300], the temperature is [20, 22, 25, 28, 30], the humidity is [60, 65, 70, 75, 80], and the rainfall is [50, 60, 70, 80, 90]. Use time series analysis methods, such as the ARIMA model, to model pest populations. The AIC criterion is used to select optimal parameters. Assuming the ARIMA (1,1,1) model provides the best fit, its coefficients are AR1 = 8 and MA1 = 6. Next, introduce environmental factors as exogenous variables into the model and conduct a multivariate time series analysis. Use the Granger causality test to determine the impact of environmental factors on pest populations. Assume that temperature, humidity, and rainfall all significantly affect pest populations, with coefficients of 5, 3, and 2, respectively. Use this model to predict the number of pests in the next three months, assuming the predicted result is [350, 400, 450]. Finally, combine the predicted values of environmental factors, such as temperature [32, 34, 36], humidity [85, 90, 95], and rainfall [100, 110, 120], and adjust the predicted value of the pest population, and the final prediction result is [360, 410, 460].

[0097] S109: Comprehensively analyze the degree of plant damage, pest density, and its changing trend, construct a pest and disease risk assessment model, perform pest and disease detection based on the pest and disease risk assessment model, and obtain detection results.

[0098] Sensors are used to acquire plant damage and pest density data to determine initial state parameters. Regression analysis is used to process these damage and density data to identify trend patterns. Based on these trend patterns, a risk assessment model is constructed to determine the pest and disease risk distribution. Pest and disease detection is performed based on this risk distribution to identify high-risk areas. If a high-risk area exceeds a preset threshold, a convolutional neural network is used to analyze the damage image and determine the specific pest and disease type. The test results are then combined with the trend patterns to update the risk assessment model parameters. The updated model is then used to detect a new round of pests and diseases, resulting in optimized detection results.

[0099] Specifically, when constructing the pest and disease risk assessment model, remote sensing technology and ground monitoring data were first used to obtain information on plant damage and pest density. For example, in a particular cornfield, 15% of the leaf area was damaged, and the pest density was 50 per square meter. Time series analysis revealed that pest density had been increasing by 5% per day over the past week. Based on this data, a multiple linear regression algorithm was used to build the model, with damage severity as the dependent variable and pest density and its changing trend as the independent variables. The resulting regression equation was Y = 3X1 + 2X2 + 5, where Y represents damage severity, X1 is pest density, and X2 is the changing trend. This model calculated a current risk score of 15, exceeding the preset threshold of 10, indicating a high pest and disease risk in the area. To further validate the model's accuracy, a random forest algorithm was trained on historical data, achieving a prediction accuracy of 90%. Based on the above models and algorithms, the system automatically triggers early warning signals and generates corresponding prevention and control recommendations, such as recommending spraying pesticides within three days at a dosage of 2 liters per hectare to control the pest density to below 30 per square meter, thereby reducing the degree of plant damage to within 10%.

[0100] On the other hand, this embodiment also provides a plant pest and disease early warning system based on plant monitoring, including:

[0101] An image segmentation module is used to perform pixel-level segmentation of plant leaf images using a convolutional neural network, dividing the image into healthy tissue areas and diseased areas, and calculating the proportion of the diseased area to the total leaf area based on the segmentation results;

[0102] The contour processing module is used to extract the contour of the lesion area through the edge detection algorithm and smooth the contour in combination with morphological operations to obtain a more accurate boundary of the lesion area;

[0103] Cluster analysis module, used to perform cluster analysis on the segmented lesion area using color features and texture features;

[0104] The health assessment module is used to establish a plant health assessment model based on image segmentation results and disease type identification results to quantify the degree of plant damage;

[0105] The data acquisition module is used to obtain environmental data and image data from multiple sensor nodes distributed in the field, and aggregate the data to the regional gateway through wireless transmission;

[0106] The data preprocessing module is used to preprocess and extract features from the aggregated heterogeneous data at the regional gateway, remove noise data and redundant information, and extract key features;

[0107] The pest detection module uses a deep learning model to detect and count pests in preprocessed images, identify pests at different developmental stages, and count the number of each type of pest.

[0108] The trend prediction module is used to model the pest population change trend based on time series analysis methods and predict the pest population density changes in the future by combining environmental factors;

[0109] The risk assessment module is used to comprehensively analyze the degree of plant damage, pest density and its changing trends, build a pest and disease risk assessment model, perform pest and disease detection based on the pest and disease risk assessment model, and obtain detection results.

[0110] On the other hand, this embodiment further provides an electronic device, comprising a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the method is implemented when the processor executes the computing program.

[0111] On the other hand, this embodiment further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program implements the method when executed by a processor.

[0112] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A plant pest and disease early warning method based on plant monitoring, characterized in that: include: A convolutional neural network is used to perform pixel-level segmentation on plant leaf images, dividing the images into healthy tissue areas and diseased areas to obtain image segmentation results. The contour of the lesion area is extracted based on the edge detection algorithm, and the contour is smoothed by combining morphological operations to obtain the boundary of the lesion area; Use color and texture features to perform cluster analysis on the segmented lesion area to distinguish different types of disease symptoms and obtain disease type recognition results; Establishing a plant health assessment model based on the image segmentation results and the disease type identification results to quantify the degree of plant damage; Acquire environmental data and image data from multiple sensor nodes distributed in the field, and aggregate the data to the regional gateway through wireless transmission; At the regional gateway, the aggregated heterogeneous data is preprocessed and feature extracted to remove noise data and redundant information to obtain preprocessed images. Using a deep learning model to detect and count pests in the preprocessed images, identify pest individuals at different developmental stages, and count the number of each type of pest; Based on the time series analysis method, the pest population change trend is modeled and the pest population density change in the future is predicted in combination with environmental factors; A pest and disease risk assessment model is constructed based on the plant damage degree and the insect population density change, and pest and disease detection is performed based on the pest and disease risk assessment model to obtain detection results.

2. The method according to claim 1, characterized in that The process of obtaining the image segmentation result includes: Perform pixel-level segmentation on plant leaf images through convolutional neural networks to obtain preliminary segmentation areas; Based on the preliminary segmented area, the healthy tissue area and the diseased area are divided by threshold judgment, and the boundaries of the two areas are determined; Obtain the number of pixels from the healthy tissue area and the lesion area, and calculate the total pixel value of each area; The pixel value of the total leaf area is extracted from the plant leaf image to obtain the overall area data; The image segmentation result is obtained based on the overall area data.

3. The method according to claim 1, characterized in that The process of obtaining the boundary of the lesion area includes: Obtaining an initial contour of the lesion area from the image segmentation result by using an edge detection algorithm to obtain first contour data; Using morphological operations to smooth the first contour data to obtain second contour data; For the second contour data, if the boundary noise is detected to exceed the preset threshold, the Gaussian filtering algorithm is used for optimization to obtain the third contour data; According to the third contour data, a coordinate set of boundary points is obtained to determine the precise boundary of the lesion area; The precise boundary of the lesion area is verified by the regional segmentation algorithm to obtain the segmented regional data; For the segmented area data, the contour tracking method is used to extract the final boundary and obtain the optimized contour data; Based on the optimized contour data, the boundary integrity is judged through geometric calculation to obtain the final lesion area boundary.

4. The method according to claim 1, wherein The process of obtaining the disease type identification result includes: Processing the boundary of the lesion area using a color feature extraction method to obtain color feature data; Using a texture feature extraction method to analyze the boundary of the lesion area to obtain texture feature data; The color feature data and texture feature data are clustered and analyzed using the K-means clustering algorithm to obtain the symptom classification results; The disease type identification result is obtained based on the symptom classification result.

5. The method according to claim 1, wherein The process of counting the number of various pests includes: Initial data is obtained by preprocessing the image, and a deep learning model is used to extract features from the image to obtain preliminary pest positioning; Based on the preliminary positioning results, the detection model is used to analyze each area to determine the distribution of individual pests; If individual pests are detected, features are extracted based on image analysis to determine the developmental stage classification; Through the stage classification results, the pest individuals at each developmental stage are obtained and the corresponding number is counted; The quantitative statistical method is used to count the individuals of each category of pests and obtain category statistical data; Extract distribution patterns from category statistics to determine complete results of pest detection; Based on the complete results, data processing techniques are used to generate structured output and obtain final statistical information.

6. The method according to claim 1, characterized in that The process of predicting insect population density changes in the future by combining environmental factors includes: Obtain historical pest population data and corresponding environmental factor data, construct a time series data set, preprocess the time series data, including data cleaning, normalization and stationarity test; use the autoregressive moving average model to model the pest population time series, and obtain preliminary prediction results; establish a multiple regression model based on the environmental factor data to analyze the impact of environmental factors on the number of pests; construct an integrated prediction model by combining the time series model and the multiple regression model; optimize the parameters of the integrated prediction model, minimize the prediction error, and improve the model accuracy; use the optimized integrated model to predict the changes in insect population density in the future.

7. The method according to claim 1, characterized in that The process of obtaining the test result includes: The plant damage degree and pest density data are acquired through sensors to determine the initial state parameters; Regression analysis was used to process the data on the degree of damage and density to obtain the characteristics of the changing trend; Build a risk assessment model based on the characteristics of the changing trend to obtain the risk distribution of pests and diseases; Conduct pest and disease detection based on risk distribution and identify high-risk areas; If the high-risk area exceeds the preset threshold, the damage level image is analyzed through a convolutional neural network to determine the specific type of pest and disease; Obtain test results and update risk assessment model parameters based on change trends; The updated model is used to detect a new round of pests and diseases and obtain optimized detection results.

8. A plant pest and disease early warning system based on plant monitoring, characterized in that: include: An image segmentation module is used to perform pixel-level segmentation of plant leaf images using a convolutional neural network, dividing the image into healthy tissue areas and diseased areas, and calculating the proportion of the diseased area to the total leaf area based on the segmentation results; The contour processing module is used to extract the contour of the lesion area through the edge detection algorithm and smooth the contour in combination with morphological operations to obtain a more accurate boundary of the lesion area; Cluster analysis module, used to perform cluster analysis on the segmented lesion area using color features and texture features; The health assessment module is used to establish a plant health assessment model based on image segmentation results and disease type identification results to quantify the degree of plant damage; The data acquisition module is used to obtain environmental data and image data from multiple sensor nodes distributed in the field, and aggregate the data to the regional gateway through wireless transmission; The data preprocessing module is used to preprocess and extract features from the aggregated heterogeneous data at the regional gateway, remove noise data and redundant information, and extract key features; The pest detection module uses a deep learning model to detect and count pests in preprocessed images, identify pests at different developmental stages, and count the number of each type of pest. The trend prediction module is used to model the pest population change trend based on time series analysis methods and predict the pest population density changes in the future by combining environmental factors; The risk assessment module is used to comprehensively analyze the degree of plant damage, pest density and its changing trends, build a pest and disease risk assessment model, perform pest and disease detection based on the pest and disease risk assessment model, and obtain detection results.

9. An electronic device comprising a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein: When the processor executes the computing program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images