Rice disease and pest early warning method and system based on image recognition

Abstract

The invention discloses a rice disease and insect pest early warning method and system based on image recognition, and particularly relates to the technical field of insect pest recognition, different areas of a rice field are comprehensively covered through multi-source high-resolution image collection, it is ensured that the image quality reaches a preset threshold value, median filtering and histogram equalization are adopted in the image preprocessing step, and the image quality is improved. Based on a geographic information system and a Kriging interpolation method, a spatial heterogeneity model of rice field disease transmission is established, a deep convolutional neural network is used for carrying out feature extraction on the image, the type, position and severity of diseases and pests are accurately recognized, the extracted features are combined with the spatial heterogeneity model, and the accuracy of the disease and pest detection is improved. A spatial autocorrelation analysis method is adopted to deeply evaluate the distribution mode of diseases in the rice field, and a support vector machine algorithm is applied to perform classification evaluation on the disease and pest risk, so that the problems of early warning lag and missing report caused by disease spatial heterogeneity are effectively solved, and the accuracy and timeliness of early warning are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of pest identification, and more specifically, to a rice pest early warning method and system based on image recognition. Background Art

[0002] As an important food crop in the world, the yield and quality of rice are directly related to food security. However, rice is often attacked by various pests and diseases during its growth, such as rice blast, sheath blight and rice planthoppers. The outbreak of these pests and diseases will not only lead to a decrease in yield, but also may cause a decrease in food quality. Therefore, timely and accurate early warning and prevention and control of rice pests and diseases are of great significance to ensure agricultural production.

[0003] At present, early warning methods for rice pests and diseases based on image recognition have gradually been applied in the agricultural field. These methods usually use image acquisition equipment to obtain images of rice fields, and combine image processing and pattern recognition technology to identify and judge the type and severity of pests and diseases. However, in the existing technology, rice pests and diseases are distributed in patches in rice fields, with high spatial heterogeneity, resulting in significant differences in the probability and severity of disease in different regions. Traditional image acquisition methods are limited by the coverage range, acquisition angle and frequency of equipment, and it is difficult to obtain high-quality images in a timely and comprehensive manner, resulting in the early warning system being unable to accurately capture the actual distribution of diseases, and may miss high-incidence areas or lag in early warning. In addition, the existing methods lack effective spatial analysis models, do not make full use of spatial statistics and geographic information system technologies to deeply analyze the spatial characteristics of diseases, and are difficult to accurately assess the risk level of different regions, and cannot provide targeted prevention and control recommendations for agricultural managers.

[0004] In order to solve the above problems, a technical solution is now provided. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a rice disease and insect pest early warning method and system based on image recognition to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The rice pest and disease early warning method based on image recognition includes the following steps:

[0008] S1: Obtain high-resolution images covering different areas of the rice field and proceed to the next step of processing after ensuring that the image quality reaches a preset threshold;

[0009] S2: De-noising, distortion correction and illumination equalization are performed on the collected images, using algorithms including median filtering and histogram equalization;

[0010] S3: Establish a spatial heterogeneity model of rice field disease transmission based on geographic information system and Kriging interpolation to clarify the disease risk level in different regions;

[0011] S4: Use deep convolutional neural network to extract features from preprocessed images and identify pest characteristics, including the type, location and severity of pests and diseases;

[0012] S5: Combine the extracted pest and disease characteristics with the spatial heterogeneity model and use the spatial autocorrelation analysis method to evaluate the distribution pattern of pests in rice fields;

[0013] S6: Based on the analysis results of the distribution pattern of diseases in rice fields, the support vector machine algorithm is used to classify and evaluate the risks of pests and diseases, and generate early warning information containing specific locations and risk levels.

[0014] In a preferred embodiment, high-resolution images covering different areas of the rice field are obtained, and the next step of processing is performed after ensuring that the image quality reaches a preset threshold, specifically:

[0015] Unmanned aerial vehicles, satellite remote sensing and ground-based cameras are used to obtain high-resolution images covering different areas of the rice fields. The image resolution, contrast and brightness are used to ensure that the image quality meets the preset threshold before proceeding to the next step of processing.

[0016] In a preferred embodiment, the collected image is subjected to denoising, distortion correction and illumination equalization processing, using algorithms including median filtering and histogram equalization, specifically including:

[0017] S201: Use median filtering algorithm to remove random noise in the image, retain image edge details, and reduce irregular interference;

[0018] S202: Using camera calibration parameters, geometric correction is performed for lens distortion to ensure the true shape and size of the image;

[0019] S203: adjusting the brightness distribution of different areas in the image through histogram equalization technology to enhance the details of dark and bright areas;

[0020] S204: Detecting artifacts and water surface reflection areas in the image, and correcting them through a reflection light filter;

[0021] S205: Setting image clarity and contrast thresholds to automatically detect whether the processed image meets the requirements of subsequent analysis.

[0022] In a preferred embodiment, a spatial heterogeneity model of rice field disease transmission is established based on geographic information system and Kriging interpolation to clarify the disease risk level in different regions, specifically including:

[0023] S301: Collect geographical data of rice fields using geographic information system, including latitude and longitude, altitude and soil type;

[0024] S302: Record disease sample data in different rice fields, and mark the geographic coordinates and disease severity of each disease point;

[0025] S303: Establish a spatial interpolation model through Kriging interpolation method to predict the disease distribution in the unobserved area of ​​the rice field: Use the collected disease sample data to construct a semivariogram; By fitting different models, select the optimal model for interpolation calculation; Use Kriging method to interpolate the unobserved area in the rice field to predict the disease distribution in the unobserved area;

[0026] S304: Use the Kriging model to calculate the spatial heterogeneity of diseases in different areas of rice fields and analyze the spatial distribution differences of diseases: Use the Moran's I index to evaluate the spatial heterogeneity of diseases: Where I is Moran's I index, N is the total number of sampling points, and w uv represents the spatial weight of positions u and v, Z(s u ) is the disease data, is the average disease data, u and v represent different sampling point indexes;

[0027] Generate a spatial distribution map of rice field diseases based on the prediction results of Kriging interpolation;

[0028] S305: Based on the results of heterogeneity analysis, the rice fields are divided into high, medium and low risk areas to form a disease risk map.

[0029] In a preferred embodiment, a deep convolutional neural network is used to extract features from the preprocessed image to identify pest and disease characteristics, which include the type, location and severity of the pest and disease, specifically including:

[0030] S401: Select a suitable deep convolutional neural network architecture, configure network layers and parameters to adapt to the task of feature extraction of pest and disease images;

[0031] S402: Arrange and annotate the image data set, perform normalization and enhancement processing, and provide it to the neural network for learning;

[0032] S403: Automatically extract key features of pests and diseases from images by training deep convolutional neural networks;

[0033] S404: Using the trained network model, identify the type, location and severity of pests and diseases in the new image, and output the identification result.

[0034] In a preferred embodiment, the extracted pest and disease characteristics are combined with the spatial heterogeneity model, and the distribution pattern of the disease in the rice field is evaluated using the spatial autocorrelation analysis method, specifically including:

[0035] S501: Fusion of extracted disease features with spatial heterogeneity model: Alignment of disease locations identified by deep convolutional neural network with rice field coordinates in GIS system; Combination of severity, type and other features of each disease point with environmental variables in spatial model to form a disease sample set with spatial features;

[0036] S502: Use Moran's I index to calculate the spatial autocorrelation of disease characteristics and assess the clustering or dispersion pattern of diseases;

[0037] S503: Analyze the clustered areas of disease distribution and identify high-risk and low-risk areas;

[0038] S504: Evaluate the disease spread trend according to the spatial distribution pattern, and optimize the disease spread prediction model based on the evaluation results.

[0039] In a preferred embodiment, based on the analysis results of the distribution pattern of diseases in rice fields, a support vector machine algorithm is used to classify and evaluate the risk of pests and diseases, and generate early warning information containing specific locations and risk levels, specifically including:

[0040] S601: combining the spatial distribution pattern of the disease with the characteristics of the disease and insect pests to form an input data set for a support vector machine;

[0041] S602: Through supervised learning, the existing disease risk data is used to train the support vector machine classification model, optimize the model parameters, and perform multi-category classification;

[0042] S603: using the trained support vector machine model, predicting the risk level of the unlabeled area, and outputting the disease risk level of each rice field area;

[0043] S604: Generate disease warning information including specific location and risk level based on the classification assessment results.

[0044] On the other hand, the present invention provides a rice pest and disease early warning system based on image recognition, including an image acquisition and verification module, an image processing and optimization module, a spatial model construction module, a feature extraction and recognition module, a spatial analysis and evaluation module, and a risk classification and early warning module;

[0045] Image acquisition and verification module: obtain high-resolution images covering different areas of the rice field and ensure that the image quality reaches the preset threshold before entering the next step of processing;

[0046] Image processing optimization module: denoise, correct distortion and balance illumination of the collected images, using algorithms including median filtering and histogram equalization;

[0047] Spatial model building module: Based on geographic information system and Kriging interpolation, a spatial heterogeneity model of rice field disease transmission is established to clarify the disease risk level in different regions;

[0048] Feature extraction and recognition module: Use deep convolutional neural network to extract features from preprocessed images and identify pest and disease characteristics, including the type, location and severity of pests and diseases;

[0049] Spatial analysis and evaluation module: Combine the extracted pest and disease characteristics with the spatial heterogeneity model and use the spatial autocorrelation analysis method to evaluate the distribution pattern of pests in rice fields;

[0050] Risk classification and early warning module: Based on the analysis results of the distribution pattern of diseases in rice fields, the support vector machine algorithm is used to classify and evaluate the risks of pests and diseases, and generate early warning information containing specific locations and risk levels.

[0051] The technical effects and advantages of the rice pest and disease early warning method and system based on image recognition of the present invention are as follows:

[0052] 1. Through multi-source high-resolution image acquisition, using drones, satellite remote sensing and ground-based cameras, we comprehensively cover different areas of the rice fields and ensure that the image quality reaches the preset threshold, overcoming the problems of limited coverage and insufficient data acquisition of traditional methods. The image preprocessing step uses median filtering and histogram equalization to improve the clarity and uniformity of the image, laying a good foundation for subsequent analysis. Based on the geographic information system and Kriging interpolation method, a spatial heterogeneity model of rice field disease transmission was established to accurately assess the disease risk level in different areas. A deep convolutional neural network was used to extract features from the image to accurately identify the type, location and severity of pests and diseases.

[0053] 2. Combine the extracted features with the spatial heterogeneity model, and use the spatial autocorrelation analysis method to deeply evaluate the distribution pattern of diseases in rice fields. Finally, the support vector machine algorithm is used to classify and evaluate the risk of pests and diseases, and generate early warning information containing specific locations and risk levels. This method effectively solves the problems of early warning lag and underreporting caused by the spatial heterogeneity of diseases, improves the accuracy and timeliness of early warning, provides agricultural managers with accurate and reliable decision-making support, and helps to take targeted prevention and control measures in a timely manner to ensure the safety and stability of rice production. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A schematic diagram of the rice pest and disease early warning method based on image recognition according to the present invention;

[0055] Figure 2 The figure is a schematic diagram of the structure of the rice disease and insect pest early warning system based on image recognition according to the present invention. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] Example 1

[0058] Figure 1 The present invention provides a rice pest and disease early warning method based on image recognition, which comprises the following steps:

[0059] S1: Obtain high-resolution images covering different areas of the rice field and ensure that the image quality reaches a preset threshold before proceeding to the next step of processing.

[0060] S2: De-noising, distortion correction and illumination equalization are performed on the acquired images, using algorithms including median filtering and histogram equalization.

[0061] S3: Establish a spatial heterogeneity model of rice field disease transmission based on geographic information system and Kriging interpolation to clarify the disease risk level in different regions.

[0062] S4: A deep convolutional neural network is used to extract features from the preprocessed images and identify pest and disease characteristics, including the type, location, and severity of pests and diseases.

[0063] S5: The extracted pest and disease characteristics were combined with the spatial heterogeneity model, and the spatial autocorrelation analysis method was used to evaluate the distribution pattern of pests and diseases in rice fields.

[0064] S6: Based on the analysis results of the distribution pattern of diseases in rice fields, the support vector machine algorithm is used to classify and evaluate the risks of pests and diseases, and generate early warning information containing specific locations and risk levels.

[0065] Obtain high-resolution images covering different areas of the rice field, and ensure that the image quality reaches the preset threshold before proceeding to the next step of processing, specifically:

[0066] Unmanned aerial vehicles, satellite remote sensing and ground-based cameras are used to obtain high-resolution images covering different areas of the rice fields. The image resolution, contrast and brightness are used to ensure that the image quality meets the preset threshold before proceeding to the next step of processing.

[0067] Select a drone with high-resolution imaging capabilities and the ability to fly stably at a specified altitude. The drone is equipped with an optical camera with a resolution of at least 12MP (12 million pixels) to ensure that it can capture subtle signs of pests and diseases in the rice fields. The flight path of the drone is controlled by a preset route planning software to ensure uniform coverage of all areas of the rice fields.

[0068] Select remote sensing satellites that provide high-resolution images with a spatial resolution of at least 30 cm / pixel. This resolution can identify disease characteristics over a large range of rice fields and is particularly suitable for overall monitoring of large-scale rice fields.

[0069] Choose a high-resolution camera with a pixel count of at least 16MP (16 million pixels) and set it at a key location in the rice field. Use a fixed bracket to ensure that the camera's viewing angle is stable to capture close-up details of pests and diseases.

[0070] To ensure the quality of images entering the subsequent analysis steps, the acquired images are subjected to strict quality control. The quality of the images is evaluated based on three criteria: resolution, contrast, and brightness.

[0071] Ensure that the collected images have sufficient pixel density to distinguish disease spots and pests on rice leaves. The image resolution of drones and ground cameras must reach 12MP or above to ensure the ability to capture details.

[0072] Under different lighting conditions, the disease symptoms in rice fields (such as yellow leaves, black spots, etc.) may be similar to the background color. To ensure the clear display of disease characteristics, an image contrast enhancement algorithm is used. The image contrast standard calculates the difference between the maximum grayscale value and the minimum grayscale value of the image and ensures that the difference is greater than a certain threshold (for example, the difference is greater than 50) to avoid losing detail information under low contrast conditions.

[0073] The brightness of the image needs to be kept within a moderate range to avoid information loss due to being too dark or too bright. Image brightness is calculated by averaging the brightness values ​​of the pixels in the image and ensuring that the brightness values ​​are within a preset range (e.g., between 50 and 200) to maintain the visual balance of the image.

[0074] During the image acquisition process, the system evaluates the quality of the acquired images in real time. If the resolution, contrast or brightness does not meet the preset threshold, the drone or ground equipment will be automatically notified to re-acquire the image until an image that meets the requirements is obtained.

[0075] The flight altitude and speed of the drone are dynamically adjusted according to the terrain and weather conditions of the rice fields. For example, in conditions with strong winds, the system will reduce the flight speed of the drone to reduce the impact of motion blur on image quality.

[0076] The collected images are subjected to denoising, distortion correction and illumination equalization processing, using algorithms including median filtering and histogram equalization, including:

[0077] S201: Use median filtering algorithm to remove random noise in the image, retain image edge details, and reduce irregular interference:

[0078] During the image acquisition process, due to changes in ambient light, equipment limitations or other external factors, noise may be generated in the image, resulting in inaccurate subsequent image analysis. In order to remove random noise in the image, the median filter algorithm is used.

[0079] Median filtering is an effective nonlinear filter, especially suitable for removing impulse noise (salt and pepper noise) in images. Its processing method is based on a window, sorting the pixel values ​​in the window by size, and replacing the pixel value in the center of the window with the median value. The advantage of median filtering is that it can not only remove noise, but also better preserve the edge details of the image, avoiding the blurring effect like mean filtering.

[0080] The window size of the median filter is set to 3×3 or 5×5, depending on the noise density. The larger the window, the stronger the filtering effect, but some image details may be lost; a smaller window can retain more edge information.

[0081] For any pixel, the following formula is used for filtering: p′ i，j =median{p i-2，j-2 , p i-2，j-1 , ..., p i+2，j+2}; where p i，j represents the pixel value at row i and column j in the original image, p′ i，j is the pixel value after filtering, median is the median, i and j represent the row and column indices of the pixel in the image respectively.

[0082] The median is defined as the value in the middle of a set of data after sorting. If the number of data is an odd number, the median is the value in the middle; if the number of data is an even number, the median is the average of the two numbers in the middle. The median filter algorithm sorts the pixel values ​​in the window and takes the median as the output value of the center pixel. This can effectively remove salt and pepper noise in the image while retaining edge details.

[0083] The processed images are verified by the noise detection algorithm. Images with noise density below 0.1 can pass this step and enter the next stage of processing.

[0084] S202: Use the camera calibration parameters to perform geometric correction for lens distortion to ensure the true shape and size of the image:

[0085] Image acquisition devices such as drone cameras or satellite remote sensing imaging devices will produce lens distortion in actual use, resulting in distortion of the image geometry. In order to restore the true proportion and shape of the image, distortion correction is required.

[0086] Lens distortion categories include: pixels on the periphery of the image expand outward, causing objects to appear more curved, usually in wide-angle lenses. Pixels at the edge of the image shrink inward, making the center appear to be enlarged.

[0087] In order to solve these distortion problems, camera calibration parameters are used for geometric correction.

[0088] Camera calibration obtains the internal and external parameters of the lens through calibration images (such as chessboard images), mainly including focal length, principal point coordinates, distortion coefficients and other parameters. With these parameters, the following model can be constructed for correction:

[0089] x′=x(1+k1r 2 +k2r 4 )+2p1xy+p2(r 2 +2x 2 );

[0090] y′=y(1+k1r 2 +k2r 4 )+p1(r 2 +2y 2 )+2p2xy.

[0091] Where x′ and y′ represent the new horizontal and vertical coordinates of the pixel in the image after distortion correction, x and y are the coordinates of the pixel in the original image, k1 and k2 are the radial distortion coefficients, p1 and p2 are the tangential distortion coefficients, and r is the distance to the principal point.

[0092] According to the calibration parameters, the coordinates of each pixel are remapped to the corrected coordinate position, ensuring that the geometric shape of the image is restored.

[0093] The distortion-corrected image is verified by testing the straightness of the grid lines. If the straight objects in the image show little bending, the distortion correction is considered effective.

[0094] S203: Use histogram equalization technology to adjust the brightness distribution of different areas in the image and enhance the details of dark and bright areas:

[0095] Due to unstable lighting conditions during shooting, differences in lighting such as shadows and bright spots in rice fields can lead to uneven brightness distribution in different areas of the image. These uneven lighting conditions will affect the extraction of pest and disease features, so lighting equalization is required. Histogram equalization is a common image enhancement technology that redistributes the brightness values ​​of pixels in an image to make the brightness distribution of the image more uniform, thereby enhancing the details of the image.

[0096] Histogram equalization maps original pixel values ​​to new pixel values ​​by calculating the cumulative distribution function of image pixel values.

[0097] After equalization, the original image and the processed image are compared to check whether the brightness distribution is more uniform. After the image details are enhanced, the characteristics of pests and diseases in the rice field can be better captured.

[0098] S204: Detect artifacts and water surface reflection areas in the image, and correct them through a reflection light filter to ensure image clarity:

[0099] Since there are a lot of water surface reflections or other interference factors in the rice field environment, artifacts or reflected light may be generated in the image, affecting the image quality. Therefore, these interferences must be detected and processed.

[0100] Artifacts usually appear as sudden highlights or low brightness areas in the image, which have a large brightness difference from the surrounding pixels. The local area brightness difference detection algorithm is used to identify these artifact areas.

[0101] By detecting the gradient of pixel brightness changes in the image, the artifact area is identified and corrected using an interpolation method. The interpolation method recalculates the pixel values ​​in the artifact area using the brightness values ​​of neighboring pixels to ensure a smooth transition.

[0102] The processed images were verified by checking the correction effect in the artifact areas to ensure that the overall brightness of the image was uniform.

[0103] S205: Set the image clarity and contrast thresholds to automatically detect whether the processed image meets the requirements of subsequent analysis:

[0104] Each image is quantitatively evaluated, including edge definition detection, contrast analysis, and brightness distribution evaluation. Edge definition uses the Sobel edge detection algorithm to calculate the edge strength of the image; contrast is calculated by counting the grayscale differences of the image.

[0105] After image preprocessing is completed, the system automatically generates an image quality score. Only images whose scores reach the preset threshold can enter subsequent analysis and processing.

[0106] A spatial heterogeneity model of rice field disease transmission was established based on geographic information system and Kriging interpolation to clarify the disease risk level in different regions, including:

[0107] S301: Collect geographical data of rice fields using geographic information system, including latitude and longitude, altitude and soil type information:

[0108] First, GIS technology is used to collect basic geographic information of rice fields. Specific data types include: longitude and latitude coordinates, rice field altitude, soil type, and regional climate information. The geographical location and longitude (latitude, longitude) of rice fields can be accurately measured by global positioning system (GPS) equipment, while the altitude information is obtained through digital elevation model (DEM). This geographic information has a direct impact on the spread of diseases. For example, altitude may affect the temperature and humidity environment of the disease, and soil type may determine the growth of rice, thereby affecting the susceptibility of the disease.

[0109] All collected geographic data are stored in vector format (VectorFormat), longitude and latitude information are expressed in point form (Point), and the boundaries of rice fields are stored in polygon form (Polygon). Altitude data and soil types are stored in raster format (RasterFormat). This can improve the spatial resolution of the data and ensure the accuracy of spatial analysis and subsequent interpolation operations.

[0110] S302: Record disease sample data in different rice fields, and mark the geographic coordinates and severity of each disease point:

[0111] In order to establish a spatial heterogeneity model, disease sample data from various areas of the rice field are collected. The specific process includes:

[0112] Sampling points are set in the rice fields, and the distance between each sampling point is a fixed value (for example, a sampling point is set every 500 meters).

[0113] The disease type, disease area and severity of the disease were recorded at each sampling point, and the precise geographic coordinates of each disease point were marked using GPS equipment.

[0114] The severity of these disease samples was classified from 0 (no disease) to 5 (severe disease) according to agronomic standards and standardized to ensure data consistency.

[0115] In order to conduct spatial analysis, the disease sample data is stored in GIS through point datasets (PointDataset), and each point sample contains its corresponding disease severity and location coordinates. At the same time, the sample data is associated with the geographical information of the rice field (such as altitude and soil type) to ensure the close connection between the disease data and the geographical environment.

[0116] S303: Establish a spatial interpolation model through Kriging interpolation method to predict the distribution of diseases in unobserved areas of rice fields:

[0117] Kriging interpolation is a spatial statistical method that infers the disease risk of unsampled areas based on known sampling point data. The advantage of Kriging interpolation is that it not only considers the distance between sampling points, but also the spatial autocorrelation between them. Spatial autocorrelation reflects the degree of association between disease samples in different regions and helps to build a global model of disease distribution in rice fields.

[0118] The mathematical model of Kriging interpolation is as follows: Among them, Z(s0) represents the interpolation prediction value, that is, the disease risk value of the unobserved area; Z(s u ) indicates the known sampling point s u Disease data; u is the weight coefficient for each sampling point, which is calculated by the semivariogram according to the spatial position of each sampling point; n represents the total number of known sampling points used to predict the unobserved area in the interpolation calculation.

[0119] The semivariogram is defined as: Among them, γ(h) is the semivariogram function, h is the distance between two sampling points, and Z(s u ) and Z(s u +h) represent the difference in disease data between two sampling points, and N(h) represents the number of sampling point pairs with a distance of h.

[0120] The semivariogram is used to characterize the disease variation trend at different distances and help determine the weight in the interpolation process.

[0121] The collected disease sample data are used to construct the semivariogram. By fitting different models (such as spherical model and exponential model), the optimal model is selected for interpolation calculation.

[0122] Kriging was used to interpolate unobserved areas in rice fields and predict the distribution of diseases in these areas.

[0123] S304: Use the Kriging model to calculate the spatial heterogeneity of diseases in different areas of rice fields and analyze the differences in the spatial distribution of diseases:

[0124] Spatial heterogeneity refers to the degree of difference in the distribution of diseases in different areas of a rice field. Since the spread of diseases in rice fields is usually not uniform, the spatial distribution of diseases needs to be analyzed. Kriging interpolation not only provides predictions, but also generates error estimates that reflect the accuracy of the interpolation model in different areas.

[0125] The spatial autocorrelation of diseases in different areas of the rice field was calculated, and the Moran's I index was used to assess the spatial heterogeneity of the disease: Where I is Moran's I index, N is the total number of sampling points (for spatial autocorrelation analysis, it means the total number of all sampling points considered in the analysis), and w uv represents the spatial weight of positions u and v, Z(s u ) is the disease data, is the average disease data, u and v represent different sampling point indexes (used to identify the spatial locations of different sampling points in the rice field).

[0126] Based on the prediction results of Kriging interpolation, the spatial distribution map of rice field diseases was generated. Based on these data, the spatial heterogeneity of different regions was calculated and the characteristics of disease transmission in different areas within the rice field were analyzed.

[0127] S305: Based on the results of heterogeneity analysis, the rice fields are divided into high, medium and low risk areas to form a disease risk map:

[0128] Based on the results of spatial heterogeneity analysis, combined with the distribution density, severity and spatial correlation of diseases, rice fields are divided into high-risk, medium-risk and low-risk areas. The specific division criteria are as follows:

[0129] High-risk areas: areas with predicted disease severity higher than 4.0 and low spatial heterogeneity, that is, areas with strong spatial concentration and diffusion of diseases.

[0130] Medium risk areas: areas with predicted disease severity between 2.0 and 4.0, with moderate heterogeneity.

[0131] Low-risk areas: areas with disease severity less than 2.0 and greater heterogeneity, that is, the disease distribution is relatively dispersed.

[0132] The disease risk level is visualized to generate a disease risk map. The disease risk map uses color coding to display the risk level of different areas, with red indicating high risk, yellow indicating medium risk, and green indicating low risk areas. Through this map, farmers and agricultural managers can intuitively understand the distribution of rice field diseases and take targeted prevention and control measures.

[0133] By combining GIS technology with Kriging interpolation, the spatial heterogeneity of rice field diseases is solved, and the disease risk distribution in unobserved areas can be effectively predicted. Through fine spatial interpolation and heterogeneity analysis, the present invention can generate accurate disease risk maps, helping to achieve timely early warning and prevention and control strategy formulation of rice field diseases.

[0134] A deep convolutional neural network is used to extract features from the preprocessed images and identify pest and disease characteristics, which include the type, location, and severity of pests and diseases, including:

[0135] S401: Select a suitable deep convolutional neural network architecture, configure network layers and parameters to adapt to the task of feature extraction of pest and disease images:

[0136] Select classic architectures based on deep convolutional neural networks, such as ResNet (residual network) or VGG (visual geometry group network) to ensure effective extraction of pest and disease image features.

[0137] The characteristic of ResNet is that it solves the gradient vanishing problem during deep network training by introducing "skip connections". For this task, ResNet-50 is selected, which contains 50 layers and can effectively capture the complex features in pest and disease images.

[0138] The VGG network uses a deeper but simpler structure, using small convolution kernels (3×3) to stack convolution layers to extract image features at different levels. In some cases, VGG is able to perform a more fine-grained analysis of the morphological characteristics of pests and diseases.

[0139] The key parameters of the network architecture include the size of the convolution kernel, the number of convolution layers, the pooling layer configuration, and the number of nodes in the fully connected layer. The specific configuration is as follows:

[0140] Convolutional layer: Each convolutional layer has a kernel size of 3×3 and uses the ReLU activation function for nonlinear processing. The number of convolutional layers depends on the depth of the network architecture (for example, ResNet-50 has 50 layers).

[0141] Pooling layer: Use maxpooling with a pooling kernel size of 2×2 to reduce the dimension of the feature map and control the model complexity.

[0142] Fully connected layer: The last layer is a fully connected layer, which is used to generate the final prediction results of pest and disease classification. The output layer is set to a number of nodes according to the number of disease types (for example, when there are five diseases, the output layer is set to 5 nodes).

[0143] The Adam optimizer is selected, and the learning rate is initially set to 0.001 to ensure the convergence of network training. At the same time, the cross-entropy loss function (Cross-EntropyLoss) is selected for classification tasks to calculate the difference between the model prediction value and the actual label.

[0144] S402: Organize and annotate the image data set, perform normalization and enhancement processing, and provide it to the neural network for learning:

[0145] Obtain rice field pest and disease image data from multiple sources. The data set needs to cover various common disease types, such as rice blast, sheath blight, and false smut. In order to ensure the diversity of the data set, the image data should include rice field disease images at different growth stages, different lighting conditions, and different viewing angles. Each image should be clearly labeled with the disease type, location, and severity of the disease.

[0146] Use annotation tools (such as LabelImg or VGGImageAnnotator) to manually annotate the images, record the specific location of each diseased area (bounding box or pixel-level segmentation) and the type and severity of the disease. The annotated dataset will be used as training data for supervised learning.

[0147] In order to improve the generalization ability of the model, the image data is normalized and the pixel values ​​are mapped to the range of [0,1] to eliminate the differences in image brightness and contrast. In addition, data enhancement is performed to improve the robustness of the model to different scenes. The data enhancement operations include:

[0148] Rotate: Randomly rotate the image by ±15 degrees.

[0149] Scaling: Randomly scale the image with a scaling factor between 0.8 and 1.2.

[0150] Flip: Randomly flip the image horizontally or vertically.

[0151] Brightness Adjustment: Randomly adjust the brightness of the image to simulate disease manifestations under different lighting conditions.

[0152] These data augmentation operations help expand the diversity of the training dataset and avoid model overfitting.

[0153] S403: Automatically extract key features of pests and diseases from images by training deep convolutional neural networks:

[0154] The normalized and enhanced image data is input into the deep convolutional neural network for training. The training process includes the following steps:

[0155] The input image is passed through convolutional layers, pooling layers, and fully connected layers to extract local and global features in the image layer by layer, and finally the type, location, and severity of pests and diseases are predicted through the output layer.

[0156] The cross entropy loss function is used to calculate the error between the model's prediction results and the true label.

[0157] The back propagation algorithm is used to adjust the weight parameters in the model to minimize the value of the loss function. The optimizer uses the Adam optimizer to update the weights at an adaptive learning rate.

[0158] The dataset is divided into a training set (80%) and a validation set (20%). After each epoch in the training process, the validation set is used to evaluate the performance of the model. By monitoring the loss value and accuracy on the validation set, hyperparameters (such as learning rate, batch size, etc.) are adjusted to ensure the generalization ability of the model on new data.

[0159] The convolutional layer can extract pest and disease features of different scales from the image, such as the morphology, color, and texture of the lesions. The convolutional neural network gradually extracts more advanced pest and disease features through layer-by-layer processing of multiple convolutional layers, thereby distinguishing different types of diseases.

[0160] S404: Using the trained network model, identify the type, location and severity of pests and diseases in the new image, and output the identification result:

[0161] After training is completed, new images are input into the deep convolutional neural network, and the model will automatically recognize them based on the learned features. The specific recognition content includes:

[0162] Disease type: Determine the type of pest and disease, such as rice blast and sheath blight, based on the activation of the network’s output layer nodes.

[0163] Disease location: The network’s feature map is combined with an object detection algorithm (such as YOLO or FasterR-CNN) to locate the specific location where the disease occurs and output the boundary box of the disease.

[0164] Disease severity: Based on the area of ​​lesions, color changes and other characteristics, the severity of the disease is assessed and divided into mild, moderate and severe.

[0165] The identification results are output in the form of structured data, including the type, location and severity of the disease. The final output can generate a visual image, annotate the diseased area, and attach identification information to help users understand the distribution and risk of rice field diseases.

[0166] The extracted pest and disease characteristics were combined with the spatial heterogeneity model, and the spatial autocorrelation analysis method was used to evaluate the distribution pattern of pests in rice fields, including:

[0167] S501: Fusion of extracted disease features with spatial heterogeneity model:

[0168] The disease locations identified by the deep convolutional neural network are aligned with the rice field coordinates in the GIS system to ensure that each disease point has an accurate location mark in the spatial heterogeneity model.

[0169] The severity and type of each disease point are combined with environmental variables (such as altitude and soil moisture) in the spatial model to form a disease sample set with spatial characteristics. These sample data will be used for subsequent spatial autocorrelation analysis.

[0170] S502: Use Moran's I index to calculate the spatial autocorrelation of disease characteristics and evaluate the clustering or dispersion pattern of diseases:

[0171] The spatial distribution matrix of the disease is constructed by combining the extracted disease characteristics (type, severity, etc.) with the spatial heterogeneity model.

[0172] Moran's I index was used to calculate global autocorrelation and determine the spatial aggregation or dispersion pattern of the disease.

[0173] The corresponding spatial weight of each disease point is calculated according to its location and the relationship with the adjacent disease points, and then the spatial autocorrelation is obtained.

[0174] S503: Analyze the clustered areas of disease distribution, identify high-risk and low-risk areas, and display them graphically:

[0175] Highly clustered areas: If the Moran's I index of a certain area is significantly positive, it means that the disease distribution in the area has a high spatial clustering and may be a hot spot for the disease. Such highly clustered areas are usually the outbreak centers of the disease and must be given priority prevention and control.

[0176] Dispersed area: If the Moran's I index is negative, it means that the disease distribution in the area is relatively dispersed and the spread of the disease may be slow, but it still needs to be monitored.

[0177] The spatial distribution pattern of diseases is visualized through the Geographic Information System (GIS) to generate a disease risk map. The risk map uses different colors to indicate the degree of disease concentration.

[0178] S504: Evaluate the disease spread trend according to the spatial distribution pattern, and optimize the disease spread prediction model based on the evaluation results:

[0179] By analyzing spatial autocorrelation and disease distribution patterns, the spread trend of the disease can be inferred. According to the aggregation of different regions, the possible spread path of the disease can be predicted. For example, if some high-risk areas are closely connected with the surrounding medium- and low-risk areas, the high-risk areas may become the spread source of the disease, and the disease has a tendency to spread to the surrounding areas.

[0180] Based on the results of spatial autocorrelation analysis, the disease spread prediction model was further optimized. By using clustered data as the model input, the model's ability to predict future disease spread paths can be improved. The optimized model can more accurately identify high-risk areas in rice fields and update disease risk assessments in real time.

[0181] Based on the analysis results of the distribution pattern of diseases in rice fields, the support vector machine algorithm is used to classify and evaluate the risks of pests and diseases, and generate early warning information containing specific locations and risk levels, including:

[0182] S601: Combine the spatial distribution pattern of the disease with the characteristics of the disease and insect pests to form the input data set of the support vector machine:

[0183] From step S4 and step S5, two main data sources have been obtained:

[0184] Disease feature data: feature information such as disease type, severity, and location. These data come from feature extraction using deep convolutional neural networks.

[0185] Spatial distribution pattern of diseases: Disease distribution pattern and aggregation index, such as Moran's I index, are obtained through spatial autocorrelation analysis.

[0186] Each disease site is assigned a type label (e.g., rice blast, sheath blight, etc.), as well as the severity of the disease. The severity can be based on area or color characteristics and can be classified as mild, moderate, or severe.

[0187] Based on the spatial location (longitude and latitude) of each disease point, the spatial relationship data of its adjacent disease points are combined to form spatial autocorrelation features. The Moran's I index is used to reflect the disease aggregation degree, and the input features are expanded by combining these spatial information.

[0188] Combining the spatial distribution pattern of the disease with its image features, multi-dimensional input features are provided for the support vector machine model. The input features of each data point include disease type, severity, spatial location, spatial autocorrelation (such as Moran's I), etc. Through these features, the global information of the disease distribution is combined with the local features to provide a complete input data set for the support vector machine classification.

[0189] S602: Through supervised learning, the existing disease risk data is used to train the support vector machine classification model, optimize the model parameters, and perform multi-category classification:

[0190] Through supervised learning, the existing disease risk data is used to train the support vector machine model. The existing disease risk data comes from historical disease samples, including risk labels (high, medium, and low risk) for each disease area. These labels are used together with the previous input features to train the support vector machine model.

[0191] Support vector machines support different kernel functions for classification tasks. Due to the nonlinear distribution of disease risk classification, the radial basis function (RBF) kernel is used to handle the complex nonlinear relationship of the input data.

[0192] The two main parameters of the support vector machine model are the penalty parameter and the kernel function parameter. Through grid search and cross-validation methods, these two parameters are optimized to ensure the accuracy and generalization ability of the classification model.

[0193] Since disease risk classification involves three risk levels (high, medium, and low), a one-vs-all strategy is adopted to transform the multi-category problem into multiple binary classification problems, thereby achieving multi-category classification.

[0194] S603: Use the trained support vector machine model to predict the risk level of the unlabeled areas and output the disease risk level of each rice field area:

[0195] The extracted disease features (disease type, severity, spatial distribution pattern, etc.) are input into the trained support vector machine model.

[0196] The model predicts the risk level of each disease area based on the input disease characteristics, and the output is a classification label of high risk, medium risk or low risk.

[0197] The forecast results for each disease area include the following information:

[0198] Disease location: The geographical location information of the GIS system provides the specific latitude and longitude coordinates of the diseased area;

[0199] Risk level: The classification result shows the risk level of the area (high risk, medium risk or low risk);

[0200] Confidence: The output of the support vector machine model can also include the confidence of the classification, which indicates how sure the model is about the prediction result. A result with a higher confidence level means that the model is more accurate in judging the risk.

[0201] S604: Generate disease warning information including specific location and risk level based on the classification assessment results, and visualize it through the GIS system or publish it to relevant personnel:

[0202] According to the risk level prediction result in S603, disease warning information including specific location and risk level is generated. The content of the warning information includes:

[0203] Disease location: Based on the geographic coordinates in the rice field, the specific location of each high, medium and low risk area is output.

[0204] Risk level: Assign corresponding risk labels to each area, clearly showing that high-risk areas need to be handled first.

[0205] Confidence and other additional information: Based on the model output, additional confidence information and a description of the disease type and severity are provided to help agricultural managers understand the context of the warning information.

[0206] The classification results of disease risk levels are displayed graphically through the Geographic Information System (GIS). The visualized disease risk map uses different colors to represent different risk areas:

[0207] Red: indicates high-risk area; yellow: indicates medium-risk area; green: indicates low-risk area.

[0208] These visualization results can help farmers and agricultural managers quickly identify key areas of disease in rice fields and develop targeted prevention and control strategies.

[0209] The generated warning information can be released through a variety of channels, including:

[0210] Email or SMS: Send warning information to relevant farmers or agricultural management departments to ensure that the warning information can be conveyed in a timely manner.

[0211] Real-time map update: The disease risk map is updated in real time on the GIS platform to ensure that users can view the latest disease risk assessment results at any time.

[0212] Example 2

[0213] The difference between Example 2 of the present invention and Example 1 is that this example introduces a rice disease and insect pest early warning system based on image recognition.

[0214] Figure 2 A structural schematic diagram of a rice disease and insect pest early warning system based on image recognition of the present invention is given. The rice disease and insect pest early warning system based on image recognition includes an image acquisition and verification module, an image processing optimization module, a spatial model construction module, a feature extraction and recognition module, a spatial analysis and evaluation module, and a risk classification and early warning module.

[0215] Image acquisition and verification module: obtain high-resolution images covering different areas of the rice field and ensure that the image quality reaches the preset threshold before entering the next step of processing.

[0216] Image processing optimization module: performs denoising, distortion correction and illumination equalization on the collected images, using algorithms including median filtering and histogram equalization.

[0217] Spatial model building module: Based on geographic information system and Kriging interpolation, a spatial heterogeneity model of rice field disease transmission is established to clarify the disease risk level in different regions.

[0218] Feature extraction and recognition module: A deep convolutional neural network is used to extract features from the preprocessed images and identify the characteristics of pests and diseases, including the type, location and severity of pests and diseases.

[0219] Spatial analysis and evaluation module: The extracted pest and disease characteristics are combined with the spatial heterogeneity model, and the spatial autocorrelation analysis method is used to evaluate the distribution pattern of pests in rice fields.

[0220] Risk classification and early warning module: Based on the analysis results of the distribution pattern of diseases in rice fields, the support vector machine algorithm is used to classify and evaluate the risks of pests and diseases, and generate early warning information containing specific locations and risk levels.

[0221] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0222] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0223] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0224] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and modules described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0225] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0226] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, and may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0227] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0228] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0229] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0230] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A rice pest and disease early warning method based on image recognition, characterized in that: The steps include: S1: Obtain high-resolution images covering different areas of the rice field and proceed to the next step of processing after ensuring that the image quality reaches a preset threshold; S2: De-noising, distortion correction and illumination equalization are performed on the collected images, using algorithms including median filtering and histogram equalization; S3: Establish a spatial heterogeneity model of rice field disease transmission based on geographic information system and Kriging interpolation to clarify the disease risk level in different regions; S4: Use deep convolutional neural network to extract features from preprocessed images and identify pest characteristics, including the type, location and severity of pests and diseases; S5: Combine the extracted pest and disease characteristics with the spatial heterogeneity model and use the spatial autocorrelation analysis method to evaluate the distribution pattern of pests in rice fields; S6: Based on the analysis results of the distribution pattern of diseases in rice fields, the support vector machine algorithm is used to classify and evaluate the risks of pests and diseases, and generate early warning information containing specific locations and risk levels.

2. The rice pest and disease early warning method based on image recognition according to claim 1, characterized in that: Obtain high-resolution images covering different areas of the rice field, and ensure that the image quality reaches the preset threshold before proceeding to the next step of processing, specifically: Unmanned aerial vehicles, satellite remote sensing and ground-based cameras are used to obtain high-resolution images covering different areas of the rice fields. The image resolution, contrast and brightness are used to ensure that the image quality meets the preset threshold before proceeding to the next step of processing.

3. The rice pest and disease early warning method based on image recognition according to claim 2, characterized in that: The collected images are subjected to denoising, distortion correction and illumination equalization processing, using algorithms including median filtering and histogram equalization, including: S201: Use median filtering algorithm to remove random noise in the image, retain image edge details, and reduce irregular interference; S202: Using camera calibration parameters, geometric correction is performed for lens distortion to ensure the true shape and size of the image; S203: adjusting the brightness distribution of different areas in the image through histogram equalization technology to enhance the details of dark and bright areas; S204: Detecting artifacts and water surface reflection areas in the image, and correcting them through a reflection light filter; S205: Setting image clarity and contrast thresholds to automatically detect whether the processed image meets the requirements of subsequent analysis.

4. The rice pest and disease early warning method based on image recognition according to claim 3, characterized in that: A spatial heterogeneity model of rice field disease transmission was established based on geographic information system and Kriging interpolation to clarify the disease risk level in different regions, including: S301: Collect geographical data of rice fields using geographic information system, including latitude and longitude, altitude and soil type; S302: Record disease sample data in different rice fields, and mark the geographic coordinates and disease severity of each disease point; S303: Establish a spatial interpolation model through Kriging interpolation method to predict the disease distribution in the unobserved area of ​​the rice field: Use the collected disease sample data to construct a semivariogram; By fitting different models, select the optimal model for interpolation calculation; Use Kriging method to interpolate the unobserved area in the rice field to predict the disease distribution in the unobserved area; S304: Use the Kriging model to calculate the spatial heterogeneity of diseases in different areas of rice fields and analyze the spatial distribution differences of diseases: Use the Moran's I index to evaluate the spatial heterogeneity of diseases: Where I is Moran's I index, N is the total number of sampling points, and w uυ represents the spatial weight of position u and υ, Z(s u ) is the disease data, is the average disease data, u and υ represent different sampling point indexes; Generate a spatial distribution map of rice field diseases based on the prediction results of Kriging interpolation; S305: Based on the results of heterogeneity analysis, the rice fields are divided into high, medium and low risk areas to form a disease risk map.

5. The rice pest and disease early warning method based on image recognition according to claim 4, characterized in that: A deep convolutional neural network is used to extract features from the preprocessed images and identify pest and disease characteristics, which include the type, location, and severity of pests and diseases, including: S401: Select a suitable deep convolutional neural network architecture, configure network layers and parameters to adapt to the task of feature extraction of pest and disease images; S402: Arrange and annotate the image data set, perform normalization and enhancement processing, and provide it to the neural network for learning; S403: Automatically extract key features of pests and diseases from images by training deep convolutional neural networks; S404: Using the trained network model, identify the type, location and severity of pests and diseases in the new image, and output the identification result.

6. The rice pest and disease early warning method based on image recognition according to claim 5, characterized in that: The extracted pest and disease characteristics were combined with the spatial heterogeneity model, and the spatial autocorrelation analysis method was used to evaluate the distribution pattern of pests in rice fields, including: S501: Fusion of extracted disease features with spatial heterogeneity model: Alignment of disease locations identified by deep convolutional neural network with rice field coordinates in GIS system; Combination of severity, type and other features of each disease point with environmental variables in spatial model to form a disease sample set with spatial features; S502: Use Moran's I index to calculate the spatial autocorrelation of disease characteristics and assess the clustering or dispersion pattern of diseases; S503: Analyze the clustered areas of disease distribution and identify high-risk and low-risk areas; S504: Evaluate the disease spread trend according to the spatial distribution pattern, and optimize the disease spread prediction model based on the evaluation results.

7. The rice pest and disease early warning method based on image recognition according to claim 6, characterized in that: Based on the analysis results of the distribution pattern of diseases in rice fields, the support vector machine algorithm is used to classify and evaluate the risks of pests and diseases, and generate early warning information containing specific locations and risk levels, including: S601: combining the spatial distribution pattern of the disease with the characteristics of the disease and insect pests to form an input data set for a support vector machine; S602: Through supervised learning, the existing disease risk data is used to train the support vector machine classification model, optimize the model parameters, and perform multi-category classification; S603: using the trained support vector machine model, predicting the risk level of the unlabeled area, and outputting the disease risk level of each rice field area; S604: Generate disease warning information including specific location and risk level based on the classification assessment results.

8. A rice pest and disease early warning system based on image recognition, used to implement the rice pest and disease early warning method based on image recognition according to any one of claims 1 to 7, characterized in that: It includes image acquisition and verification module, image processing and optimization module, space model building module, feature extraction and recognition module, space analysis and evaluation module and risk classification and early warning module; Image acquisition and verification module: obtain high-resolution images covering different areas of the rice field and ensure that the image quality reaches the preset threshold before entering the next step of processing; Image processing optimization module: denoise, correct distortion and balance illumination of the collected images, using algorithms including median filtering and histogram equalization; Spatial model building module: Based on geographic information system and Kriging interpolation, a spatial heterogeneity model of rice field disease transmission is established to clarify the disease risk level in different regions; Feature extraction and recognition module: Use deep convolutional neural network to extract features from preprocessed images and identify pest and disease characteristics, including the type, location and severity of pests and diseases; Spatial analysis and evaluation module: Combine the extracted pest and disease characteristics with the spatial heterogeneity model and use the spatial autocorrelation analysis method to evaluate the distribution pattern of pests in rice fields; Risk classification and early warning module: Based on the analysis results of the distribution pattern of diseases in rice fields, the support vector machine algorithm is used to classify and evaluate the risks of pests and diseases, and generate early warning information containing specific locations and risk levels.

Images