A pepper disease and pest automatic identification method and system suitable for a field

Abstract

This invention provides an automatic identification method and system for chili pepper diseases and pests adapted for field use, belonging to the field of agricultural disease and pest monitoring technology. The technical solution includes: collecting and preprocessing hyperspectral images of chili pepper disease and pest samples to obtain disease and pest categories and evaluation parameter labels; constructing a dual-branch hybrid spectral-spatial depth network model and training it using a multi-task loss function to obtain a disease and pest identification model; dividing the chili pepper field into multiple sub-regions and obtaining sub-region observation vectors through coarse acquisition of hyperspectral images; triggering fine acquisition of hyperspectral images of suspected diseased areas based on Mahalanobis distance calculation and adaptive trigger threshold judgment, and inputting the data into the trained identification model to obtain disease and pest categories and evaluation parameters; constructing a disease and pest evaluation index, combined with a risk threshold range set for the chili pepper growth stage, to complete disease and pest identification and risk rating. This invention solves the problems of low efficiency and poor accuracy of traditional identification methods in field use.

Description

Technical Field

[0001] This invention relates to the field of agricultural pest and disease monitoring technology, specifically to an automatic identification method and system for chili pepper pests and diseases adapted for field use. Background Technology

[0002] In the context of large-scale modern agricultural cultivation, chili peppers, as a key economic crop, are directly affected by pest and disease control, which directly impacts final yield and product quality. Traditional pest and disease monitoring methods rely heavily on regular field inspections by agricultural technicians and their experience-based identification. This approach is not only inefficient but also prone to omissions and misjudgments, making it unsuitable for the real-time monitoring requirements of large-scale field cultivation. In recent years, with the rapid advancements in spectral imaging and artificial intelligence technologies, technologies utilizing hyperspectral images for intelligent identification of crop pests and diseases have gradually become a research focus in this field.

[0003] Among existing technologies, CN116883835A discloses an automatic identification method for agricultural product diseases based on a local-global collaborative network. This method uses hyperspectral imaging technology to acquire images of agricultural product samples and achieves automatic disease identification through a local-global collaborative network model. CN115019215A proposes a soybean pest and disease identification method based on hyperspectral images, utilizing a drone equipped with a hyperspectral camera to collect data and combining it with a course-based learning approach for model training. CN114550108A discloses an identification and early warning method for fall armyworm, constructing a pest identification model by extracting spectral and texture features and generating early warning information based on the degree of damage. CN120259789A provides a multi-variety citrus Huanglongbing monitoring method based on hyperspectral imaging, selecting core feature bands and constructing a discriminative model using a convolutional neural network. The prior art disclosed in CN113468964A is a method for monitoring agricultural pests and diseases based on hyperspectral imaging, which identifies the type and level of pests and diseases by counting the number of damaged leaves and combining them with a neural network model.

[0004] However, existing technologies still have the following shortcomings: First, current automatic identification methods lack adaptive data collection schemes that fit the actual field environment. Most adopt fixed data collection methods and cannot dynamically adjust the monitoring frequency and scope according to the actual field conditions, resulting in data redundancy and wasting a lot of computing resources. Second, existing technologies have not yet established an effective quantitative assessment system for the degree of pest and disease damage, making it difficult to accurately determine the occurrence stage and spatial distribution density of pests and diseases, and failing to provide a refined risk classification basis for agricultural production. Third, while ensuring identification accuracy, existing identification algorithms cannot simultaneously meet the efficiency requirements of large-scale field monitoring. Especially in large-scale planting areas, conducting full-coverage, high-frequency monitoring would bring enormous data processing pressure. Finally, existing technologies do not have a dynamic threshold adjustment mechanism that matches the different growth stages of chili peppers, and cannot automatically optimize monitoring parameters as the crop growth status changes, thus reducing the accuracy of identification results and the effectiveness of practical applications. Summary of the Invention

[0005] The purpose of this invention is to provide an automatic identification method and system for chili pepper diseases and pests adapted to field use, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: An automatic identification method for pepper diseases and pests adapted for field use, comprising the following steps: We collected hyperspectral images of chili pepper plant samples showing morphological changes caused by pests and diseases, as well as hyperspectral images of healthy chili pepper plant samples at various growth stages. We then labeled the pest and disease categories and chili pepper pest and disease evaluation parameters corresponding to the hyperspectral images of the pest and disease samples. A deep network model was built, with hyperspectral images of pest and disease samples as input, and the corresponding pest and disease categories and pepper pest and disease evaluation parameters as labels. The model was trained to obtain a pepper pest and disease identification model. Band analysis was performed on the hyperspectral images of healthy samples to obtain the health spectral index corresponding to each growth stage. Based on the health spectral index, the observation vector of healthy peppers at each growth stage was obtained, and a spectral benchmark library of healthy peppers was built. The chili field was divided into multiple sub-regions. A fixed acquisition time interval was set, and hyperspectral images were coarsely acquired for each sub-region to obtain the first image of each sub-region. Band analysis was performed on the first image to obtain the observation vector of the sub-region. Based on the observation vectors of sub-regions and growth stages, the Mahalanobis distance between the observation vectors of sub-regions and the spectral reference library of healthy peppers is calculated. An adaptive trigger threshold is set. When the threshold is greater than the adaptive trigger threshold, the morphological characteristics of pepper plants in the sub-region are acquired by high-spectral image acquisition. Based on detailed collection of chili plant morphological characteristics within a sub-region, hyperspectral images of suspected lesions are obtained. These images are then input into a trained pest and disease identification model to obtain chili pest and disease evaluation parameters and pest and disease category information. A chili pest and disease evaluation index is constructed based on chili pest and disease evaluation parameters. The risk threshold range of different pest and disease evaluation indices is defined according to the growth stage of the chili. Based on the matching results of the chili pest and disease evaluation index and the risk threshold range, the identification and rating of pests and diseases in each sub-region are completed.

[0007] Furthermore, hyperspectral images of leaf and fruit morphological characteristics caused by pests and diseases in chili peppers were collected. These pests and diseases included those susceptible to occur at various growth stages of chili peppers. For each pest and disease, hyperspectral images of samples at each infection stage were collected. The pre-processed hyperspectral images of pest and disease samples were manually delineated to obtain the diseased areas. Pest and disease category labels and pest and disease evaluation parameter labels were then added. The pest and disease evaluation parameters included: pest and disease severity, pest and disease development level, and pest and disease spatial distribution density.

[0008] Furthermore, the system labels pest and disease categories and evaluation parameter labels are generated. One-hot encoding is used to label pests and diseases, and the evaluation parameter labels are three-dimensional vectors. The specific formulas involved are as follows: in, Labels for pest and disease evaluation parameters. The severity of pests and diseases. To manually delineate the area of ​​diseased pixels within the diseased region of hyperspectral images of pest and disease samples. The total pixel area of ​​the sample hyperspectral image. As the degree of development of pests and diseases, This refers to the spatial distribution density of pests and diseases. To manually delineate the number of independent lesions within the susceptible area of ​​hyperspectral images of pest and disease samples. The number of visible insects.

[0009] Furthermore, a deep network model was constructed, employing a dual-branch hybrid spectral-spatial network architecture. Branch 1: The average spectral vector extracted from the susceptible region of a manually delineated hyperspectral image of a pest sample is input, passes through two fully connected layers, and is processed using the ReLU activation function to output spectral features. Branch 2 inputs the feature map of the diseased region after PCA dimensionality reduction from a manually delineated hyperspectral image of a pest and disease sample. It uses three 3×3 two-dimensional convolutional blocks with 32, 64, and 128 channels respectively. Each convolutional block is followed by batch normalization and ReLU activation, connected to a spectral attention module. Global average pooling is used to obtain the importance weight vector for each band, which is then multiplied band-by-band. The weighted cube is input into the convolutional layer, and global average pooling is used to obtain the spatial-spectral features. ;right and By splicing the data, the fused features are obtained. The system outputs pest and disease category predictions and pepper pest and disease evaluation parameters through two parallel fully connected layers; a multi-task loss function is defined. The specific formula is as follows: in, This represents the weighting coefficient for pest and disease categories. These are the weighting coefficients for pest and disease evaluation parameters; For cross-entropy loss, For mean square error loss, Predict probability distributions for pest and disease categories. This is a predicted value for the severity of pests and diseases. To predict the probability distribution of the development level of pests and diseases. The predicted values ​​are the distribution density of pests and diseases; the initial learning rate is 0.001, the batch size is 32, and the model weights with the minimum loss on the validation set are saved as the final pepper pest and disease identification model.

[0010] Furthermore, band analysis was performed on the hyperspectral images of healthy samples to obtain the health spectral indices corresponding to each growth stage. These health spectral indices include the modified normalized vegetation index (NDI) and the disease spectral sensitivity index, with the specific formulas as follows: in, To improve the normalized vegetation index, The spectrum sensitivity index of the disease. These are the weighting coefficients. The average reflectance in the near-infrared band. The average reflectance in the green light band 1, The average reflectance of the red light band 1. The average reflectance of the blue light band 1. The average reflectance of the red band 2. The average reflectance in the red band 3 is... The average reflectance of the green light band 2. The average reflectance in the blue light band 2; a spectral benchmark library for healthy chili peppers is established, using the following formula: in, Growth stage Healthy pepper observation vector at time, For those in the growth stage Healthy chili peppers average value, For those in the growth stage Healthy chili peppers average value, For the spectral reference library of healthy chili peppers Mid-growth stage The health benchmark model, For healthy chili peppers during their growth stage The covariance matrix.

[0011] Furthermore, the chili fields were divided into For each sub-region, a fixed acquisition time interval is set, and hyperspectral images are coarsely acquired to obtain the first image of each sub-region. The first image is a hyperspectral image of the planted field area in the sub-region. Band analysis is performed on the first image to obtain the observation vector of the sub-region, as shown in the following formula: in, sub-region The observation vector, For coarse collection of sub-regions of average value, For coarse collection of sub-regions of Average value; when the Mahalanobis distance between the observation vector of the sub-region and the healthy pepper model reaches the adaptive trigger threshold, high-spectral images of the pepper plant morphology characteristics within the sub-region are collected; the growth stage of the observation vector is calculated relative to the growth stage in the healthy pepper spectral reference library. The specific formula for the Mahalanobis distance between the healthy pepper models is as follows: in, sub-region Observation vectors and growth stages in the spectral model of healthy peppers The Mahalanobis distance between the observation vectors of healthy chili peppers at a given time; setting an adaptive trigger threshold; the adaptive trigger threshold is dynamically configured based on the chili pepper growth stage and a pre-stored healthy chili pepper spectral model; the observation vector and the growth stage in the healthy chili pepper spectral model are... The Mahalanobis distance between healthy chili pepper models follows a chi-square distribution, and an adaptive trigger threshold is used. That is, the degrees of freedom are Chi-square distribution If the quantile is greater than the adaptive trigger threshold, then the morphological characteristics of chili plants in the sub-region will be finely acquired using hyperspectral images.

[0012] Furthermore, the morphological characteristics of chili plants within the sub-region are further refined using hyperspectral images. Specifically, this refined acquisition involves: initially locating suspected diseased plants within the sub-region using color and texture features; acquiring hyperspectral images of suspected lesions at a fixed resolution using a hyperspectral imager for the upper and middle parts of the plant canopy; inputting these suspected lesion hyperspectral images as independent samples into the chili pest and disease identification model, which outputs the pest and disease category and evaluation parameters corresponding to each suspected lesion hyperspectral image; calculating the most frequently occurring pest and disease category in the sub-region as the primary pest and disease, and using the weighted average of all pest and disease evaluation parameters output by the chili pest and disease identification model from the refined acquisition of lesion hyperspectral images within the sub-region as the representative value of the sub-region.

[0013] Furthermore, an evaluation index for chili pepper diseases and pests is constructed based on the evaluation parameters, using the following formula: in, sub-region Evaluation index of pepper diseases and pests. Main diseases and pests The corresponding weight vector, sub-region The weighted average of the severity of pests and diseases. sub-region The weighted average of the degree of development of pests and diseases. sub-region The weighted average of the spatial distribution density of pests and diseases. Main diseases and pests The corresponding weighting coefficients for pest and disease severity, pest and disease development level, and pest and disease spatial distribution density satisfy the following conditions: and In the [0.1] interval.

[0014] Furthermore, risk threshold ranges for different pest and disease assessment indices are defined based on the growth stage of the chili peppers. Specifically: for each growth stage... Define three risk thresholds ,Will The value range is divided into four risk intervals; when Time region This is a low-risk area for pests and diseases. Time region This area is considered a medium-risk area for pests and diseases. Time region Areas at high risk of pests and diseases, when Time region This area is considered an area of ​​immediate risk for pests and diseases.

[0015] The present invention further provides an automatic identification system for chili pepper diseases and pests adapted to field use, and an automatic identification method for chili pepper diseases and pests adapted to field use for performing any of the above-mentioned methods, comprising: Data acquisition and annotation module: This module is used to collect hyperspectral images of chili plant samples showing morphological changes caused by pests and diseases, as well as hyperspectral images of healthy chili plant samples at various growth stages. It also annotates the pest and disease categories and chili pest and disease evaluation parameters corresponding to the hyperspectral images of the pest and disease samples. The identification model training module is used to build a deep network model. It takes hyperspectral images of pest and disease samples as input, and the corresponding pest and disease categories and pepper pest and disease evaluation parameters as labels to train the model and obtain a pepper pest and disease identification model. The Healthy Benchmark Library Construction Module is used to perform band analysis on the hyperspectral images of healthy samples, obtain the health spectral index corresponding to each growth stage, obtain the observation vector of healthy peppers at each growth stage based on the health spectral index, and build a healthy pepper spectral benchmark library. Sub-region coarse acquisition module: used to divide the chili field into multiple sub-regions, set a fixed acquisition time interval, perform coarse acquisition of hyperspectral images of each sub-region to obtain the first image of each sub-region, perform band analysis on the first image to obtain the sub-region observation vector; Adaptive inspection trigger module: Based on the sub-region observation vector and growth stage, it calculates the Mahalanobis distance between the sub-region observation vector and the healthy pepper spectral reference library, sets an adaptive trigger threshold, and when it is greater than the adaptive trigger threshold, it performs high-spectral image acquisition of the morphological characteristics of pepper plants in the sub-region. The identification and analysis module is used to collect detailed images of chili plant morphology characteristics within a sub-region, obtain hyperspectral images of suspected lesions, input these images into a trained pest and disease identification model, and obtain chili pest and disease evaluation parameters and pest and disease category information. The comprehensive evaluation output module is used to construct a chili disease and pest evaluation index based on chili disease and pest evaluation parameters. It defines the risk threshold range of different disease and pest evaluation indices according to the growth stage of the chili. Based on the matching results of the chili disease and pest evaluation index and the risk threshold range, it completes the identification and rating of diseases and pests in each sub-region.

[0016] Compared with existing technologies, the beneficial effects of this invention are as follows: By combining coarse sampling and inspection with adaptive fine sampling, the efficiency of field pest and disease monitoring is significantly improved, avoiding data redundancy caused by high-resolution sampling across the entire area; the anomaly monitoring mechanism using Mahalanobis distance and adaptive thresholds can accurately locate suspected disease areas, reducing invalid data collection compared to traditional fixed sampling modes; a three-dimensional evaluation parameter system including severity, development level, and spatial distribution density is established, enabling quantitative evaluation of pest and disease severity and providing a scientific basis for agricultural production decisions; through a dual-branch deep network and spectral attention module, the spectral and spatial information of hyperspectral data is fully utilized, improving identification accuracy compared to single-feature methods; the constructed risk grading system can dynamically adjust evaluation standards according to different growth stages of chili peppers, achieving personalized pest and disease risk assessment. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall method flow of the present invention; Figure 2 This is a diagram of experimental data for a sub-region in an embodiment of the present invention; Figure 3 This is a block diagram of the system module structure of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0019] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0020] Example: Please see Figures 1-2 The present invention provides a technical solution: An automatic identification method for pepper diseases and pests adapted for field use, comprising the following steps: We collected hyperspectral images of chili pepper plant samples showing morphological changes caused by pests and diseases, as well as hyperspectral images of healthy chili pepper plant samples at various growth stages. We then labeled the pest and disease categories and chili pepper pest and disease evaluation parameters corresponding to the hyperspectral images of the pest and disease samples. In this embodiment, hyperspectral images of leaf and fruit morphological characteristics caused by pests and diseases affecting chili peppers are collected. These pests and diseases include those prevalent at various growth stages of chili peppers, including anthracnose, blight, powdery mildew, scab, aphids, thrips, and mites. For each pest and disease, hyperspectral images of samples from the initial, middle, and late infection stages are included. Each category contains at least 200 individual samples, resulting in an image library of at least 1600 samples. The hyperspectral images of the pest and disease samples are preprocessed. Specifically, the preprocessing steps include: performing dark current correction and whiteboard correction on the data cubes in the original hyperspectral images; and finally, normalizing the standard score of the spectral curve of each pixel. The preprocessed hyperspectral images of the pest and disease samples are then manually delineated to obtain the manually delineated disease-affected areas. These disease-affected areas are defined as regions containing typical symptoms and with clear boundaries, and are labeled with pest and disease category tags. Labels for pest and disease evaluation parameters The evaluation parameters for pests and diseases include: severity of pests and diseases, degree of development of pests and diseases, and spatial distribution density of pests and diseases.

[0021] Labeling pest and disease categories and evaluation parameters involves using one-hot encoding for pest and disease labeling, and the evaluation parameter labels are three-dimensional vectors. The specific formulas are as follows: in, Labels for pest and disease evaluation parameters. The severity of pests and diseases. To manually delineate the area of ​​diseased pixels within the diseased region of hyperspectral images of pest and disease samples. The total pixel area of ​​the sample hyperspectral image. As the degree of development of pests and diseases, This refers to the spatial distribution density of pests and diseases. To manually delineate the number of independent lesions within the susceptible area of ​​hyperspectral images of pest and disease samples. The number of visible insects; for To quantify and classify the levels of development Directly label it as a discrete level, Normalize to the [0,1] interval.

[0022] Severity of pests and diseases It directly reflects the proportion of leaf or fruit area that has lost green color or become necrotic, and is a key indicator for measuring yield loss potential. and It can be calculated directly from image pixels, making it objective and repeatable. Pixel area is chosen over actual area because in a calibrated image, the pixel ratio and the actual area ratio have a fixed relationship, and the calculation is more convenient. Increase or Reduction will lead to Increased indicates a more serious condition.

[0023] Disease and pest development level Based on characteristics such as lesion color, texture, and presence of spores, agronomic experts classify lesions into discrete grades. This is an empirical parameter used to characterize the temporal dimension of the disease; the spatial distribution density of pests and diseases. middle To delineate the number of individual lesions within the area, This represents the theoretical maximum number of patches in the entire image, or, for pests, the number of visible insects. This value reflects whether the disease is sporadic or widespread. Under the same circumstances, The larger the lesion, the more fragmented it is. The higher the value, the more severe the prevention and control situation.

[0024] This step explicitly requires samples to cover all growth and infection stages, resolving misclassification issues caused by changes in growth stages or different disease development phases in practical applications, and improving the model's generalization ability and robustness. Existing technologies often only collect samples for specific periods or severely symptomatic diseases. Simultaneously, by manually and meticulously delineating infected areas and labeling 3D evaluation parameter vectors, high-quality supervision signals beyond simple classification are provided for subsequent training of multi-task deep learning models. This enables the model not only to identify categories but also to learn the quantitative characteristics of disease severity. It provides source data and definition standards for the subsequent calculation of pest and disease evaluation indices.

[0025] A deep network model was built, with hyperspectral images of pest and disease samples as input, and the corresponding pest and disease categories and pepper pest and disease evaluation parameters as labels. The model was trained to obtain a pepper pest and disease identification model. In this embodiment, a deep network model is constructed. The model adopts a dual-branch hybrid spectral-spatial network architecture. Branch 1: The average spectral vector extracted from the diseased area of ​​the hyperspectral image of manually delineated pest and disease samples is input, processed through two fully connected layers, and then processed using the ReLU activation function to output spectral features. Branch 2 inputs the feature map of the diseased region after PCA dimensionality reduction from a manually delineated hyperspectral image of a pest and disease sample. It uses three 3×3 two-dimensional convolutional blocks with 32, 64, and 128 channels respectively. Each convolutional block is followed by batch normalization and ReLU activation, connected to a spectral attention module. Global average pooling is used to obtain the importance weight vector for each band, which is then multiplied band-by-band. The weighted cube is input into the convolutional layer, and global average pooling is used to obtain the spatial-spectral features. ;right and By splicing the data, the fused features are obtained. The system outputs pest and disease category predictions and pepper pest and disease evaluation parameters through two parallel fully connected layers; a multi-task loss function is defined. The specific formula is as follows: in, This represents the weighting coefficient for pest and disease categories. These are the weighting coefficients for pest and disease evaluation parameters; For cross-entropy loss, For mean square error loss, Predict probability distributions for pest and disease categories. This is a predicted value for the severity of pests and diseases. To predict the probability distribution of the development level of pests and diseases. The predicted values ​​are the distribution density of pests and diseases; the initial learning rate is 0.001, the batch size is 32, and the model weights with the minimum loss on the validation set are saved as the final pepper pest and disease identification model.

[0026] In this embodiment, the input is the average spectral vector extracted from the diseased region, processed by two fully connected layers, and the output is spectral features. The feature map is obtained after PCA dimensionality reduction, retaining the first 10 principal components. The selection of the first 10 principal components is based on a balance between empirical analysis and computational efficiency: typically, the first few principal components of hyperspectral data can capture over 90% of the total variance, and 10 principal components are sufficient to retain key spectral features related to plant physiological state and stress. Simultaneously, reducing the number of input channels from 128 to 10 significantly reduces the computational burden and memory usage of subsequent convolutional networks, improving the speed and stability of model training and inference. The input is the feature map after PCA dimensionality reduction; three 3×3 two-dimensional convolutional blocks are used, with 32, 64, and 128 channels respectively. Each convolutional block is followed by batch normalization and ReLU activation, then connected to a spectral attention module. Since the input is a 10-channel PCA feature map instead of the original 128 bands, the original SAM module needs adjustment. The modified SAM module obtains the importance weight vector for each PCA feature channel through global average pooling. The PCA feature map is multiplied channel-by-channel by the weight vector, and the weighted feature map is input into subsequent convolutional layers. This modification focuses the attention mechanism on the most informative PCA components, rather than the original bands. Finally, spatial-spectral features are obtained through a global average pooling layer. .

[0027] spectral features Spatial-spectral characteristics By concatenating the features along the feature dimension, we obtain the fused features. The output is achieved through two parallel fully connected layers: pest and disease category prediction and evaluation parameter prediction.

[0028] Define a multi-task loss function Training is conducted, and the weighting coefficients for pest and disease categories are adjusted. Pest and disease evaluation parameter weighting coefficient The Adam optimizer was used with an initial learning rate of 0.001 and a batch size of 32. The model was trained for 1000 epochs on the training set and its performance was monitored on the validation set. The model weights with the smallest loss on the validation set were saved as the final chili pepper disease and pest identification model. The test set was used for the final performance evaluation.

[0029] This step employs a dual-branch hybrid network to extract spectral features sensitive to disease and spatial features characterizing lesion shape and texture, and innovatively introduces a spectral attention module. This captures the essential information of hyperspectral data—spectral curve morphology and spatial distribution—better than simply using CNNs or networks with full-spectrum input, and has a stronger ability to identify early and atypical disease symptoms. The design of the multi-task loss function forces the model to understand deeper features related to disease severity and development while learning classification; these features, in turn, enhance classification accuracy. This is an effective regularization that improves the model's generalization performance. This model maps high-dimensional hyperspectral images to precise categories and quantization parameters, forming the foundation for all subsequent automated analyses.

[0030] Band analysis was performed on the hyperspectral images of healthy samples to obtain the health spectral index corresponding to each growth stage. Based on the health spectral index, the observation vector of healthy peppers at each growth stage was obtained, and a spectral benchmark library of healthy peppers was built. Band analysis was performed on the hyperspectral images of healthy samples to obtain the health spectral indices corresponding to each growth stage. These health spectral indices include the modified normalized vegetation index (NDI) and the disease spectral sensitivity index, with the specific formulas as follows: in, To improve the normalized vegetation index, The spectrum sensitivity index of the disease. These are the weighting coefficients. The average reflectance in the near-infrared band. The average reflectance in the green light band 1, The average reflectance of the red light band 1. The average reflectance of the blue light band 1. The average reflectance of the red light band 2. The average reflectance in the red band 3 is... The average reflectance of the green light band 2. The average reflectance in the blue light band 2; a spectral benchmark library for healthy chili peppers is established, using the following formula: in, Growth stage Healthy pepper observation vector at time, For those in the growth stage Healthy chili peppers average value, For those in the growth stage Healthy chili peppers average value, For the spectral reference library of healthy chili peppers Mid-growth stage The health benchmark model For healthy chili peppers during their growth stage The covariance matrix.

[0031] In this embodiment, ; The average reflectance at a wavelength of 800 nm. The average reflectance at a wavelength of 550 nm. The average reflectance at a wavelength of 680 nm; Average reflectance at a wavelength of 450 nm The average reflectance at a wavelength of 750 nm. The average reflectance at a wavelength of 705 nm. The average reflectance at a wavelength of 550 nm. The value represents the average reflectance at a wavelength of 450 nm. In this embodiment, a pepper planting experimental field during the flowering stage was divided into 15 sub-regions. Average and The average values ​​are shown in Table 1 below: Table 1: Average values ​​of improved normalized vegetation index and disease spectral sensitivity index for sub-regions Table 1 reflects the continuous variation spectrum of vegetation health status in regions 2, 5, 7, and 11. A high value, greater than 0.682, indicates that the plants in the corresponding sub-region are very healthy, with dark green leaves and vigorous photosynthesis; regions 1, 9, and 13... Values ​​in the range of 0.63-0.68 indicate that the plants in the corresponding sub-regions are generally healthy, but may be under mild environmental stress; values ​​in regions 3, 6, 10, and 15... Values ​​in the range of 0.58-0.63 indicate mild stress in the plants within the corresponding sub-regions, possibly due to early-stage disease or mild water or nutrient deficiencies; regions 4, 8, 12, and 14... A value less than 0.5 indicates significant plant abnormalities within the corresponding sub-region, with a severe decrease in vegetation cover or chlorophyll content; similarly, for The values ​​also reflect the different sub-regions of pepper plants, specifically regions 1, 2, 5, 7, 9, and 11. A value less than 0.13 indicates that the red edges of the pepper plants in the sub-region are normal and the leaf structure is intact; values ​​in regions 10, 13, and 15... Values ​​in the range of 0.13-0.189 indicate the beginning of a blue shift from red to blue, suggesting potential changes in cell structure and signaling early disease development or physiological stress; regions 3, 4, 6, 8, 12, and 14... A value greater than 0.18 indicates significant red edge displacement and damage to mesophyll cells. (Through...) and Two spectral indices enable rapid monitoring of 15 areas across the entire field at low cost. The data objectively reflects the continuous spectrum of pepper plants from health to disease stress, providing a comprehensive and quantitative raw data foundation for subsequent intelligent triggering of fine diagnosis, and serving as the starting point for achieving efficient and precise plant protection.

[0032] This step divides the field into sub-regions for management, aligning with the zoning management principles of precision agriculture. Regular, low-intensity data collection at fixed time intervals enables routine, low-intensity monitoring of farmland health. and These two lightweight indices, used as observation vectors, offer fast computation speeds and low hardware requirements, making them suitable for rapid inspection on UAVs. Compared to using full-band data, this reduces the burden of real-time data transmission and processing. Design rationale: Introducing the blue band corrects for the effects of atmospheric and soil background, especially with a smaller saturation effect in dense vegetation, allowing for more sensitive reflection of canopy physiological status. The red band absorbs chlorophyll, the near-infrared band reflects plant structure, and the blue band is used for atmospheric aerosol correction. The weighting design is optimized based on extensive experimental data. Design rationale: The red edge is the region in the vegetation reflectance spectrum that is most sensitive to disease stress. Diseases usually cause a blue shift in the red edge, that is, a shift towards shorter wavelengths. This ratio effectively captures subtle changes in the position of the red edge, subtracting... This item aims to eliminate changes in green and blue light caused by nutrient status or mild water stress, thereby enhancing the specificity for diseases. When disease occurs, the shape of the reflectance curve in the red-edged area changes, directly affecting the value of the first item; simultaneously, leaf chlorosis may lead to… Change; well-designed It should deviate significantly from the healthy value when the disease occurs.

[0033] The chili field was divided into multiple sub-regions. A fixed acquisition time interval was set, and hyperspectral images were coarsely acquired for each sub-region to obtain the first image of each sub-region. Band analysis was performed on the first image to obtain the observation vector of the sub-region. In this embodiment, the chili field is divided into... For each sub-region, a fixed acquisition time interval is set, and hyperspectral images are coarsely acquired to obtain the first image of each sub-region. The first image is a hyperspectral image of the planted field area in the sub-region, and band analysis is performed on the first image to obtain the observation vector of the sub-region, as shown in the following formula: in, sub-region The observation vector, For coarse collection of sub-regions of average value, For coarse collection of sub-regions of average value; Based on the observation vectors of sub-regions and growth stages, the Mahalanobis distance between the observation vectors of sub-regions and the spectral reference library of healthy peppers is calculated. An adaptive trigger threshold is set. When the threshold is greater than the adaptive trigger threshold, the morphological characteristics of pepper plants in the sub-region are acquired by high-spectral image acquisition. In this embodiment, when the Mahalanobis distance between the observation vector of a sub-region and the healthy pepper model reaches the adaptive trigger threshold, high-spectral images of the pepper plant morphology characteristics within the sub-region are acquired; the observation vector and the growth stage data in the healthy pepper spectral reference library are calculated. The specific formula for the Mahalanobis distance between the healthy pepper models is as follows: in, sub-region Observation vectors and growth stages in the spectral model of healthy peppers The Mahalanobis distance between the observation vectors of healthy chili peppers at a given time; setting an adaptive trigger threshold; the adaptive trigger threshold is dynamically configured based on the chili pepper growth stage and a pre-stored healthy chili pepper spectral model; the observation vector and the growth stage in the healthy chili pepper spectral model are... The Mahalanobis distance between healthy chili pepper models follows a chi-square distribution, and an adaptive trigger threshold is used. That is, the degrees of freedom are Chi-square distribution If the quantile is greater than the adaptive trigger threshold, then the morphological characteristics of chili plants in the sub-region will be finely acquired using hyperspectral images.

[0034] In this embodiment, a healthy pepper model during the flowering period is used as an example: , , ; , , That is, the 95th percentile of the chi-square distribution with 2 degrees of freedom is approximately 5.991; If a significant abnormality is found in a sub-region, a detailed data collection instruction is generated, and the sub-region is marked as a suspected infected area.

[0035] Euclidean distance cannot account for the correlation between features, nor the differences in variance among features; while and Typically correlated, Mahalanobis distance is standardized by introducing the inverse of the covariance matrix, eliminating the influence of dimensions and correlation, and can truly reflect the statistical distance of sample points in a multivariate distribution. It is learned entirely from health data and objectively defines the scope of health, in which... It determines the shape of the healthy range. and The correlation is high; the healthy region appears as a tilted ellipse in a two-dimensional plane. Mahalanobis distance accurately measures the distance to the center of this ellipse. Even if a point deviates only slightly from one metric, if it deviates from this correlation, for example... High but If the value is abnormally low, the Mahalanobis distance may also be very large.

[0036] This provides a statistically rigorous criterion for judging outliers in Mahalanobis distance. The sensitivity of the system was controlled. The larger the threshold The larger the size, the more conservative the system becomes. Only more significant anomalies will trigger fine-grained data collection, increasing the risk of missed reports but reducing false alarms. The smaller the value, the more sensitive the system. This embodiment calculates the Mahalanobis distance for 15 sub-regions as shown in Table 2 below: Table 2: Data Table for Mahalanobis Distance Calculation Please see Figure 2 As shown in Table 2, the healthy areas can be identified. Generally very low, especially in regions 1, 2, 5, 7, 9, and 11. The value is between 0.016 and 0.619, far below the threshold of 5.991; the diseased area Significantly elevated; regions 4, 6, 8, 12, and 14 All values ​​between 6.516 and 31.161 exceeded the threshold. Regions 3 and 15... Approaching but not exceeding the threshold of 5.991; not triggering fine-grained sampling is economical. If the disease does exist but is minor, in the next inspection 3 days later, if the disease has progressed, It will continue to rise and trigger monitoring; if it is only a temporary stress, It may return to normal.

[0037] This step introduces Mahalanobis distance and dynamic chi-square thresholding, avoiding reliance on fixed thresholds and instead using statistical distribution data from healthy populations for anomaly monitoring. It considers the natural variations in the spectral characteristics of peppers at different growth stages, reducing false triggers. It also enables on-demand data acquisition; resource-intensive fine-grained data acquisition is only initiated when a sub-region exhibits a statistically significant anomaly. This saves energy and improves system efficiency.

[0038] Based on detailed collection of chili plant morphological characteristics within a sub-region, hyperspectral images of suspected lesions are obtained. These images are then input into a trained pest and disease identification model to obtain chili pest and disease evaluation parameters and pest and disease category information. In this embodiment, the morphological characteristics of chili plants within the sub-region are acquired using hyperspectral images. Specifically, this acquisition involves: initially locating suspected diseased plants within the sub-region using color and texture features; directly facing the upper part of the plant canopy; and acquiring suspected lesion hyperspectral images at a fixed resolution using a hyperspectral imager. The suspected lesion hyperspectral images undergo the same preprocessing as in the previous steps. These preprocessed images are then input as independent samples into the chili pest and disease identification model, which outputs the pest and disease category and evaluation parameters corresponding to each suspected lesion hyperspectral image. The most frequently occurring pest and disease category in the sub-region is calculated as the primary pest and disease, and the weighted average of all pest and disease evaluation parameters output by the chili pest and disease identification model from the carefully acquired hyperspectral images within the sub-region is used as the representative value for that sub-region.

[0039] In this embodiment, the 15 sub-regions Sub-regions exceeding the adaptive trigger threshold are filtered, as shown in Table 3 below: Table 3: Fine-grained recognition results Table 2 shows that regions 4, 8, and 14 are identified as epidemic diseases, and regions 6 and 12 are identified as anthrax. If... and Quantitative grading can also reflect The severity of pests and diseases is 0.22, indicating a mild condition. The severity of pests and diseases is 0.28, indicating a moderate level. : 0.35 indicates a moderate to severe level of pest and disease severity; The severity of pests and diseases is 0.62, indicating a severe condition. : 0.71 indicates an extremely severe level of pest and disease severity; taking region 8 as an example, the average value of the three pest and disease evaluation parameters reflects that the region has been fully exposed to severe disease and needs to be dealt with immediately; in region 6, the disease is in the development stage and there is a window for prevention and control.

[0040] This step, after triggering detailed data collection, does not involve blindly taking pictures, but first performs preliminary positioning based on color and texture features. This further focuses on the target suspected diseased plants, ensuring the quality and relevance of the detailed images and avoiding the generation of invalid data.

[0041] A chili pest and disease evaluation index is constructed based on chili pest and disease evaluation parameters. The risk threshold range of different pest and disease evaluation indices is defined according to the growth stage of the chili. Based on the matching results of the chili pest and disease evaluation index and the risk threshold range, the identification and rating of pests and diseases in each sub-region are completed.

[0042] In this embodiment, a chili pest and disease evaluation index is constructed based on chili pest and disease evaluation parameters, and the formula is as follows: in, sub-region Evaluation index of pepper diseases and pests. Main diseases and pests The corresponding weight vector, sub-region The weighted average of the severity of pests and diseases. sub-region The weighted average of the degree of development of pests and diseases. sub-region The weighted average of the spatial distribution density of pests and diseases. Main diseases and pests The corresponding weighting coefficients for pest and disease severity, pest and disease development level, and pest and disease spatial distribution density satisfy the following conditions: and In the [0.1] interval.

[0043] Taking anthracnose and blight as examples of the pests and diseases involved, the corresponding weighting coefficients for pest and disease severity, pest and disease development level, and pest and disease spatial distribution density are as follows: The index calculation and risk rating are shown in Table 4 below: Table 4: PDI Index Calculation and Risk Rating Table In this embodiment, risk threshold ranges for different pest and disease assessment indices are defined according to the growth stage of the chili pepper. Specifically: for each growth stage... Define three risk thresholds ,Will The value range is divided into four risk intervals; This area is considered a low-risk area for pests and diseases. This area is considered a medium-risk area for pests and diseases. This area is at high risk of pests and diseases. This area is designated as a high-risk zone for pests and diseases. In this embodiment, , , .

[0044] This step constructs a weighted comprehensive evaluation index. This method integrates three different evaluation parameters into a single, easily understandable indicator. Setting dynamic risk thresholds based on growth stages reflects the timeliness and scientific rigor of plant protection decisions. The final output is an intuitive risk level map, rather than a collection of technical parameters, greatly facilitating understanding and decision-making for farm managers. A single parameter cannot comprehensively assess the damage. A single large lesion indicates the severity of the pest or disease. High but spatial distribution density of pests and diseases Low; multiple small lesions, severity of pests and diseases. Medium but with moderate spatial distribution density of pests and diseases Different pests and diseases require different control strategies. Weighted fusion can comprehensively reflect the intensity, duration, and spread potential of the damage. Different pests and diseases have different damage patterns, thus requiring different weight combinations; this reflects the flexibility and professionalism of the solution.

[0045] Please see Figure 3 This invention provides an automatic identification system for chili pepper diseases and pests adapted for field use. The system is used to execute the aforementioned automatic identification method for chili pepper diseases and pests adapted for field use, comprising: Data acquisition and annotation module: This module is used to collect hyperspectral images of chili plant samples showing morphological changes caused by pests and diseases, as well as hyperspectral images of healthy chili plant samples at various growth stages. It also annotates the pest and disease categories and chili pest and disease evaluation parameters corresponding to the hyperspectral images of the pest and disease samples. The identification model training module is used to build a deep network model. It takes hyperspectral images of pest and disease samples as input, and the corresponding pest and disease categories and pepper pest and disease evaluation parameters as labels to train the model and obtain a pepper pest and disease identification model. The Healthy Benchmark Library Construction Module is used to perform band analysis on the hyperspectral images of healthy samples, obtain the health spectral index corresponding to each growth stage, obtain the observation vector of healthy peppers at each growth stage based on the health spectral index, and build a healthy pepper spectral benchmark library. Sub-region coarse acquisition module: used to divide the chili field into multiple sub-regions, set a fixed acquisition time interval, perform coarse acquisition of hyperspectral images of each sub-region to obtain the first image of each sub-region, perform band analysis on the first image to obtain the sub-region observation vector; Adaptive inspection trigger module: Based on the sub-region observation vector and growth stage, it calculates the Mahalanobis distance between the sub-region observation vector and the healthy pepper spectral reference library, sets an adaptive trigger threshold, and when it is greater than the adaptive trigger threshold, it performs high-spectral image acquisition of the morphological characteristics of pepper plants in the sub-region. The identification and analysis module is used to collect detailed images of chili plant morphology characteristics within a sub-region, obtain hyperspectral images of suspected lesions, input these images into a trained pest and disease identification model, and obtain chili pest and disease evaluation parameters and pest and disease category information. The comprehensive evaluation output module is used to construct a chili disease and pest evaluation index based on chili disease and pest evaluation parameters. It defines the risk threshold range of different disease and pest evaluation indices according to the growth stage of the chili. Based on the matching results of the chili disease and pest evaluation index and the risk threshold range, it completes the identification and rating of diseases and pests in each sub-region.

[0046] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0047] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0048] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0049] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. An automatic identification method for pepper diseases and pests adapted for field use, characterized in that, The specific steps include: We collected hyperspectral images of chili pepper plant samples showing morphological changes caused by pests and diseases, as well as hyperspectral images of healthy chili pepper plant samples at various growth stages. We then labeled the pest and disease categories and chili pepper pest and disease evaluation parameters corresponding to the hyperspectral images of the pest and disease samples. A deep network model was built, with hyperspectral images of pest and disease samples as input, and the corresponding pest and disease categories and pepper pest and disease evaluation parameters as labels. The model was trained to obtain a pepper pest and disease identification model. Band analysis was performed on the hyperspectral images of healthy samples to obtain the health spectral index corresponding to each growth stage. Based on the health spectral index, the observation vector of healthy peppers at each growth stage was obtained, and a spectral benchmark library of healthy peppers was built. The chili field was divided into multiple sub-regions. A fixed acquisition time interval was set, and hyperspectral images were coarsely acquired for each sub-region to obtain the first image of each sub-region. Band analysis was performed on the first image to obtain the observation vector of the sub-region. Based on the observation vectors of sub-regions and growth stages, the Mahalanobis distance between the observation vectors of sub-regions and the spectral reference library of healthy peppers is calculated. An adaptive trigger threshold is set. When the threshold is greater than the adaptive trigger threshold, the morphological characteristics of pepper plants in the sub-region are acquired by high-spectral image acquisition. Based on detailed collection of chili plant morphological characteristics within a sub-region, hyperspectral images of suspected lesions are obtained. These images are then input into a trained pest and disease identification model to obtain chili pest and disease evaluation parameters and pest and disease category information. A chili pest and disease evaluation index is constructed based on chili pest and disease evaluation parameters. The risk threshold range of different pest and disease evaluation indices is defined according to the growth stage of the chili. Based on the matching results of the chili pest and disease evaluation index and the risk threshold range, the identification and rating of pests and diseases in each sub-region are completed.

2. The automatic identification method for pepper diseases and pests adapted to field use as described in claim 1, characterized in that: Hyperspectral images of leaf and fruit morphological characteristics caused by pests and diseases in chili peppers were collected. The pests and diseases included those susceptible to occur at various growth stages of chili peppers. For each pest and disease, hyperspectral images of samples at various infection stages were collected. The pre-processed hyperspectral images of pest and disease samples were manually delineated to obtain the diseased areas. Pest and disease category labels and pest and disease evaluation parameter labels were added. The pest and disease evaluation parameters included: pest and disease severity, pest and disease development level, and pest and disease spatial distribution density.

3. The automatic identification method for pepper diseases and pests adapted to field use as described in claim 2, characterized in that: Labeling pest and disease categories and evaluation parameters involves using one-hot encoding for pest and disease labeling, and the evaluation parameter labels are three-dimensional vectors. The specific formulas are as follows: in, Labels for pest and disease evaluation parameters. The severity of pests and diseases. To manually delineate the area of ​​diseased pixels within the diseased region of hyperspectral images of pest and disease samples. The total pixel area of ​​the sample hyperspectral image. As the degree of development of pests and diseases, This refers to the spatial distribution density of pests and diseases. To manually delineate the number of independent lesions within the susceptible area of ​​hyperspectral images of pest and disease samples. The number of visible insects.

4. The automatic identification method for pepper diseases and pests adapted to field use according to claim 2, characterized in that: A deep network model is constructed, employing a dual-branch hybrid spectral-spatial network architecture. Branch 1 takes the average spectral vector extracted from the susceptible region of a manually delineated hyperspectral image of a pest or disease sample as input, processes it through two fully connected layers using the ReLU activation function, and outputs spectral features. Branch 2 inputs the feature map of the diseased region after PCA dimensionality reduction from a manually delineated hyperspectral image of a pest and disease sample. It uses three 3×3 two-dimensional convolutional blocks with 32, 64, and 128 channels respectively. Each convolutional block is followed by batch normalization and ReLU activation, connected to a spectral attention module. Global average pooling is used to obtain the importance weight vector for each band, which is then multiplied band-by-band. The weighted cube is input into the convolutional layer, and global average pooling is used to obtain the spatial-spectral features. ;right and By splicing the data, the fused features are obtained. The system outputs pest and disease category predictions and pepper pest and disease evaluation parameters through two parallel fully connected layers; a multi-task loss function is defined. The specific formula is as follows: in, This represents the weighting coefficient for pest and disease categories. These are the weighting coefficients for pest and disease evaluation parameters; For cross-entropy loss, For mean square error loss, Predict probability distributions for pest and disease categories. This is a predicted value for the severity of pests and diseases. To predict the probability distribution of the development level of pests and diseases. The predicted values ​​are the distribution density of pests and diseases; the initial learning rate is 0.001, the batch size is 32, and the model weights with the minimum loss on the validation set are saved as the final pepper pest and disease identification model.

5. The automatic identification method for pepper diseases and pests adapted to field use according to claim 3, characterized in that: Band analysis was performed on the hyperspectral images of healthy samples to obtain the health spectral indices corresponding to each growth stage. These health spectral indices include the modified normalized vegetation index (NDI) and the disease spectral sensitivity index, with the specific formulas as follows: in, To improve the normalized vegetation index, The spectrum sensitivity index of the disease. These are the weighting coefficients. The average reflectance in the near-infrared band. The average reflectance in the green light band 1, The average reflectance of the red light band 1. The average reflectance of the blue light band 1. The average reflectance of the red light band 2. The average reflectance in the red band 3 is... The average reflectance of the green light band 2. The average reflectance in the blue light band 2; a spectral benchmark library for healthy chili peppers is established, using the following formula: in, Growth stage Healthy pepper observation vector at time, For those in the growth stage Healthy chili peppers average value, For those in the growth stage Healthy chili peppers average value, For the spectral reference library of healthy chili peppers Mid-growth stage The health benchmark model, For healthy chili peppers during their growth stage The covariance matrix.

6. The automatic identification method for pepper diseases and pests adapted to field use according to claim 5, characterized in that: Divide the chili fields into For each sub-region, a fixed acquisition time interval is set, and hyperspectral images are coarsely acquired to obtain the first image of each sub-region. The first image is a hyperspectral image of the planted field area in the sub-region. Band analysis is performed on the first image to obtain the observation vector of the sub-region, as shown in the following formula: in, sub-region The observation vector, For coarse collection of sub-regions of average value, For coarse collection of sub-regions of Average value; when the Mahalanobis distance between the observation vector of the sub-region and the healthy pepper model reaches the adaptive trigger threshold, high-spectral images of the pepper plant morphology characteristics within the sub-region are collected; the growth stage of the observation vector is calculated relative to the growth stage in the healthy pepper spectral reference library. The specific formula for the Mahalanobis distance between the healthy pepper models is as follows: in, sub-region Observation vectors and growth stages in the spectral model of healthy peppers The Mahalanobis distance between the observation vectors of healthy chili peppers at a given time; setting an adaptive trigger threshold; the adaptive trigger threshold is dynamically configured based on the chili pepper growth stage and a pre-stored healthy chili pepper spectral model; the observation vector and the growth stage in the healthy chili pepper spectral model are... The Mahalanobis distance between healthy chili pepper models follows a chi-square distribution, and an adaptive trigger threshold is used. That is, the degrees of freedom are Chi-square distribution If the quantile is greater than the adaptive trigger threshold, then the morphological characteristics of the chili plants in the sub-region will be finely acquired using hyperspectral images.

7. The automatic identification method for pepper diseases and pests adapted to field use according to claim 6, characterized in that: The process involves detailed hyperspectral image acquisition of the morphological characteristics of chili plants within a sub-region. Specifically, this involves: initially locating suspected diseased plants within the sub-region using color and texture features; acquiring hyperspectral images of suspected lesions at a fixed resolution using a hyperspectral imager, targeting the upper and middle parts of the plant canopy; inputting these suspected lesion hyperspectral images as independent samples into the chili pest and disease identification model, which outputs the pest and disease category and evaluation parameters corresponding to each suspected lesion hyperspectral image; calculating the most frequently occurring pest and disease category in the sub-region as the primary pest and disease, and using the weighted average of all pest and disease evaluation parameters output by the chili pest and disease identification model from the detailed acquisition of lesion hyperspectral images within the sub-region as the representative value of the sub-region.

8. The automatic identification method for pepper diseases and pests adapted to field use according to claim 7, characterized in that: A chili pepper disease and pest evaluation index is constructed based on chili pepper disease and pest evaluation parameters, and the formula is as follows: in, sub-region Evaluation index of pepper diseases and pests. Main diseases and pests The corresponding weight vector, sub-region The weighted average of the severity of pests and diseases. sub-region The weighted average of the degree of development of pests and diseases. sub-region The weighted average of the spatial distribution density of pests and diseases. Main diseases and pests The corresponding weighting coefficients for pest and disease severity, pest and disease development level, and pest and disease spatial distribution density satisfy the following conditions: and In the [0.1] interval.

9. The automatic identification method for pepper diseases and pests adapted to field use as described in claim 8, characterized in that: Based on the growth stage of the chili pepper, different risk threshold ranges for pest and disease assessment indices are defined. Specifically: for each growth stage... Define three risk thresholds Will The value range is divided into four risk intervals; when Time region This is a low-risk area for pests and diseases. Time region This area is considered a medium-risk area for pests and diseases. Time region Areas at high risk of pests and diseases, when Time region This area is considered an area of ​​immediate risk for pests and diseases.

10. An automatic identification system for pepper diseases and pests adapted for field use, characterized in that: An automatic identification system for chili pepper diseases and pests adapted for field use is used to execute the automatic identification method for chili pepper diseases and pests adapted for field use as described in any one of claims 1-9, comprising: Data acquisition and annotation module: This module is used to collect hyperspectral images of chili plant samples showing morphological changes caused by pests and diseases, as well as hyperspectral images of healthy chili plant samples at various growth stages. It also annotates the pest and disease categories and chili pest and disease evaluation parameters corresponding to the hyperspectral images of the pest and disease samples. The identification model training module is used to build a deep network model. It takes hyperspectral images of pest and disease samples as input, and the corresponding pest and disease categories and pepper pest and disease evaluation parameters as labels to train the model and obtain a pepper pest and disease identification model. The Healthy Benchmark Library Construction Module is used to perform band analysis on the hyperspectral images of healthy samples, obtain the health spectral index corresponding to each growth stage, obtain the observation vector of healthy peppers at each growth stage based on the health spectral index, and build a healthy pepper spectral benchmark library. Sub-region coarse acquisition module: used to divide the chili field into multiple sub-regions, set a fixed acquisition time interval, perform coarse acquisition of hyperspectral images of each sub-region to obtain the first image of each sub-region, perform band analysis on the first image to obtain the sub-region observation vector; Adaptive inspection trigger module: Based on the sub-region observation vector and growth stage, it calculates the Mahalanobis distance between the sub-region observation vector and the healthy pepper spectral reference library, sets an adaptive trigger threshold, and when it is greater than the adaptive trigger threshold, it performs high-spectral image acquisition of the morphological characteristics of pepper plants in the sub-region. The identification and analysis module is used to collect detailed images of chili plant morphology characteristics within a sub-region, obtain hyperspectral images of suspected lesions, input these images into a trained pest and disease identification model, and obtain chili pest and disease evaluation parameters and pest and disease category information. The comprehensive evaluation output module is used to construct a chili disease and pest evaluation index based on chili disease and pest evaluation parameters. It defines the risk threshold range of different disease and pest evaluation indices according to the growth stage of the chili. Based on the matching results of the chili disease and pest evaluation index and the risk threshold range, it completes the identification and rating of diseases and pests in each sub-region.

Images