Rice spectrum stress index construction method, system and equipment and storage medium

Abstract

The invention discloses a rice spectrum stress index construction method and system, computer equipment and a storage medium, and relates to the technical field of rice stress classification, and the method comprises the steps: regularly shooting different groups of rice leaves to obtain original hyperspectral images, and carrying out black and white correction; extracting regions of interest from rice leaves under different stresses, calculating average spectral data and performing SG smoothing filter processing; using a generative adversarial network to extract 50 characteristic wave bands with the highest contribution degree; inputting the average spectral data corresponding to the characteristic wave band as model input data into a random forest classification model to obtain a model classification result; and meanwhile, screening the first five characteristic wave bands and corresponding weights to construct spectral indexes, obtaining stress indexes, dividing stress categories, and outputting stress reasons of the rice. According to the method, the nonlinear relationship in the data is more efficiently captured, the classification model performance is improved, the calculation efficiency is high, the adaptability is high, the dimensionality of the hyperspectral data is greatly reduced, and the model calculation complexity is effectively reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of rice stress classification, and in particular relates to a method, system, equipment and storage medium for constructing a rice spectral stress index. Background Art

[0002] Rice is subject to a variety of biological and abiotic stresses during its growth, such as drought, fertilizer damage, and disease stress, which may affect its yield and quality. The impact mechanisms and manifestations of these stresses vary. In actual field management, the symptoms of different stresses are often difficult to distinguish. Therefore, accurate analysis of the stress factors to which rice is subjected is of great significance in actual planting and production.

[0003] Traditional rice stress diagnosis methods can be divided into field observation, laboratory testing or fixed-band vegetation index calculation. Field observation mainly relies on experienced agricultural technicians to observe and evaluate rice plants, but this method is highly subjective and easily affected by the observer's experience level and inconsistent judgment criteria, resulting in biased diagnostic results. It is also difficult to detect slight stress or small changes in the early stages of stress. Often, judgments can only be made after the symptoms of stress appear, and it is easy to miss the best time for intervention. Although laboratory testing (such as PCR diagnosis) can provide accurate diagnostic results, its operating steps are complicated. The detection process usually requires sample collection and a series of chemical treatments, which is not only time-consuming and labor-intensive, but also requires expensive professional instruments and technical support. In addition, since laboratory testing usually requires destructive sampling, it is impossible to achieve continuous monitoring of the same plant, limiting its application value in stress tracking research. In the field environment, vegetation indices are often used to assess the health of rice in stress assessment, such as the Normalized Difference Vegetation Index (NDVI). NDVI is calculated by the reflectance of red light and near-infrared bands and is often used to monitor vegetation distribution, changes and health status; the Ratio Vegetation Stress Index (RVSI) assesses the degree of vegetation stress by comparing the reflectance of red light and near-infrared bands. A high value often reflects that the plant is in a state of water stress or disease infection. Although vegetation indices such as NDVI and RVSI can reflect the overall health of the plant, these indices use a small number of bands and are not strongly correlated with the spectral response of rice under different stresses, making it difficult to distinguish the type of stress.

[0004] In recent years, models trained based on RGB images have made progress in rice stress analysis, but it is still difficult to deeply analyze the specific stress causes that affect rice health based on RGB images alone, which is still a major challenge in current research and practical applications. The stress information obtained by RGB images is limited and confined to the surface of leaves, while hyperspectral images provide rich spectral information and can obtain information that is not limited to the surface of leaves, such as changes in the content of substances inside leaves. It can capture the subtle spectral characteristics of rice leaves under different stresses, which helps to explore sensitive bands under different stresses and achieve more accurate stress analysis.

[0005] The current methods for extracting characteristic bands of rice leaves under different stresses and constructing stress indexes still have some shortcomings, mainly including the following aspects:

[0006] 1. Inaccurate selection of characteristic bands: Traditional methods usually rely on general vegetation indices or existing experience when extracting characteristic bands of rice diseases, and lack a comprehensive analysis of rice spectral data under different stresses. This results in the inability of the extracted characteristic bands to accurately characterize the typical spectral characteristics of rice under different stresses, which in turn affects the accuracy of the analysis.

[0007] 2. Lack of multi-band information fusion: The spectral response of rice involves multiple sensitive bands, and the index calculated by relying on a small number of bands cannot fully reflect the stress of rice. At the same time, the fusion effect of multi-band information in existing methods is limited, and the correlation between the bands of disease hyperspectral images cannot be fully explored through deep learning and other technologies, which limits the reliability of stress analysis.

[0008] The existing vegetation indices lack specificity and cannot accurately reflect the stress characteristics of rice: the current vegetation indices are mostly based on general indicators and fail to fully combine the unique spectral response characteristics of rice under different stresses. When identifying stress categories, the specificity and sensitivity of vegetation indices are low and cannot be directly applied. Summary of the invention

[0009] In view of the above-mentioned problems, the present invention is proposed.

[0010] Therefore, the technical problem solved by the present invention is to improve the accuracy of feature band selection. This patent implements a data-driven feature band selection method through a generative adversarial network. The generative adversarial network can automatically learn and extract the characteristic spectral bands of rice under different stresses from a large amount of hyperspectral data, effectively solving the problem of insufficient extraction accuracy caused by relying on experience or general vegetation index in traditional methods. This method can accurately capture the key spectral change characteristics of rice under disease stress, laying a solid foundation for the construction of stress index.

[0011] To achieve efficient fusion of multi-band information, the present invention proposes a multi-band fusion strategy based on generative adversarial network to address the problem that the spectral response of rice under stress involves multiple sensitive bands, but the existing methods lack fusion of multi-band information. The generator extracts and fuses the information of multiple sensitive bands in the hyperspectral data, and the discriminator evaluates the effectiveness of the fusion features, realizing dynamic mining and optimization of potential correlations between bands, overcoming the limitations of low band utilization and incomplete information expression in traditional methods. The generated features can fully reflect the spectral response characteristics of rice under stress such as drought, fertilizer damage, and disease, and improve the sensitivity and specificity of the stress index and the accuracy and reliability of stress diagnosis.

[0012] To develop a more targeted rice stress index, the present invention optimizes the band weights in the generative adversarial network based on accurate extraction of characteristic bands and fusion of multi-band information, selects the top five bands with the best effect on the stress index, and obtains the key bands and band coefficients of the stress index. Through the optimization iteration of the generative adversarial network, the stress index finally obtained can fully reflect the response and changes of rice under various stresses, has higher sensitivity and specificity, can effectively distinguish between stress types such as drought, fertilizer damage, and disease, and meet the needs of refined stress diagnosis.

[0013] To solve the above technical problems, the present invention provides the following technical solutions: a method for constructing a rice spectral stress index, comprising: using a portable hyperspectral camera to regularly photograph rice leaves of different groups on site to obtain original hyperspectral images; performing black and white correction on the original hyperspectral images to obtain corrected hyperspectral images; extracting regions of interest for rice leaves under different stresses, and calculating average spectral data; performing SG smoothing filter processing on the average spectral data; using a generative adversarial network to extract 50 feature bands with the highest contribution; inputting the average spectral data corresponding to the feature bands as model input data into a random forest classification model to obtain a model classification result; simultaneously screening the top 5 feature bands and corresponding weights to construct a spectral index, and obtaining a stress index to divide stress categories; using the model classification results and the stress category to determine the group to which the leaf belongs, and outputting the cause of rice stress.

[0014] As a preferred embodiment of the method for constructing the rice spectral stress index of the present invention, the original hyperspectral image includes:

[0015] A portable hyperspectral camera was used, natural sunlight was used as the light source, and a white board with a fixed reflectivity of 0.9 was used as a reference;

[0016] Obtain spectral information of rice leaves through live photography;

[0017] Dynamically adjust the light signal of the hyperspectral camera to maintain at 3000-4000;

[0018] Data collection was conducted in a field environment, with regular photography of healthy, drought-stressed, fertilizer-stressed, and disease-stressed rice leaves in the rice experimental field;

[0019] Recording hyperspectral image data of rice leaves under different stresses;

[0020] The rice leaf parts with obvious disease response in the group were selected as typical samples;

[0021] The resolution of the captured hyperspectral image is 1920×1920.

[0022] As a preferred embodiment of the method for constructing the rice spectral stress index of the present invention, the black and white correction includes:

[0023] The hyperspectral image is corrected by using the black-and-white correction method to obtain a corrected hyperspectral image;

[0024] Calculate the image reflectance value after black and white correction : , in, is the reflectance value of the original image to be corrected, is the reflectivity value obtained by scanning the standard reference white plate, is the blackboard reflectivity value obtained after the lens cover is put on.

[0025] As a preferred embodiment of the method for constructing the rice spectral stress index of the present invention, the extraction of the region of interest includes:

[0026] ENVI software was used to screen the areas with the most significant correlation with stress factors from the corrected hyperspectral image data and define them as areas of interest;

[0027] When rice leaves were under drought stress, fertilizer stress, or disease stress, the region of interest size was set to 50 × 50 pixels, and the average spectral data of each region was extracted;

[0028] According to the shooting period and the actually observed stress symptoms, n ROIs were marked to obtain the average spectral data records of rice leaves under n specific stresses;

[0029] When the rice leaves were healthy, the region of interest was randomly sampled using the five-point method, n ROIs were also marked, and the corresponding average spectral data were extracted.

[0030] As a preferred embodiment of the method for constructing a rice spectral stress index according to the present invention, the generative adversarial network comprises:

[0031] Generator, Discriminator;

[0032] The generator includes a fully connected layer, a batch normalization layer, a random inactivation layer and a CBAM module;

[0033] The initial input of the generator is part of the real data and a random noise vector, which is expanded to 512 dimensions through the first layer of fully connected network, and then undergoes batch normalization and random dropout with a ratio of 0.3. The second layer of fully connected network expands the features to 1024 dimensions, and then undergoes batch normalization and random dropout again. The third layer of fully connected network further expands the features to 2048 dimensions, and then undergoes batch normalization and random dropout again.

[0034] The features are reshaped into a three-dimensional tensor of (16, 16, 8), and the spatial attention and channel attention of the features are optimized through the CBAM module. The features are converted into one-dimensional vectors through the Flatten layer, and a 306-dimensional output vector is generated through the fully connected layer, representing the reflectance band value of the target hyperspectral data;

[0035] The discriminator includes using a 3D convolutional network combined with an attention mechanism to judge the input data;

[0036] The input data is reshaped into a three-dimensional data block, and three-dimensional spatial features are extracted through two sets of 3D convolutional layers and 3D maximum pooling layers;

[0037] The first set of 3D convolutional layers contains 32 filters, and the second set of 3D convolutional layers contains 64 filters, and the kernel size is (3, 3, 3);

[0038] After the CBAM module enhances the perception of feature weights, the data passes through two sets of 2D convolutional layers and 2D maximum pooling layers. After the second set of convolutions, the CBAM module is introduced again to strengthen the attention of the features, and an inactivation layer is added. The features are integrated through the flattening layer and passed to the fully connected layer to output the discrimination results.

[0039] As a preferred solution of the method for constructing the rice spectral stress index of the present invention, the 50 characteristic bands with the highest contribution are extracted, including:

[0040] Analyze the activation map of the discriminator at different convolutional layers, combine the gradient of the convolutional layer weights, locate the sensitivity of the convolution kernel to different bands, and extract the selected feature bands and corresponding weights;

[0041] After obtaining the feature weights in the convolution layer, the bands with high weight values ​​are screened out, and then the sensitive bands with the strongest correlation with drought stress, fertilizer stress, and disease stress are screened again;

[0042] By sorting and screening the weights of the sensitive feature bands output by the convolutional layer, the top 50 bands with the most contributions from 400nm to 1200nm are determined, and the spectral reflectance data corresponding to the current 50 bands are retained as the input of the model.

[0043] As a preferred embodiment of the method for constructing the rice spectral stress index of the present invention, the model classification result includes:

[0044] The dataset after characteristic band screening was used as the input of the random forest model, and the average spectral dataset was labeled according to different stress groups, including 0 for health, 1 for drought stress, 2 for fertilizer stress, and 3 for disease stress;

[0045] The random forest model includes a training phase and a prediction phase;

[0046] The training phase includes random forest randomly sampling from the training data set using the bagging method to generate m data subsets containing partial samples;

[0047] Build a decision tree on each subset, and randomly select a subset of features to split during the node partitioning process of each tree;

[0048] The prediction stage includes using all constructed decision trees to predict the test samples using random forests, and determining the final category result by majority voting;

[0049] The output of the random forest model is represented as: , in, For the final prediction result, is the total number of decision trees in the random forest, For the The prediction results of each tree;

[0050] The output of the random forest model is the corresponding stress group.

[0051] As a preferred embodiment of the method for constructing the rice spectral stress index of the present invention, the stress classification includes:

[0052] The five bands with the most positive response of rice leaves under stress were selected and sorted according to the weight values ​​of the characteristic bands, and the corresponding coefficients of the bands were obtained from the weights. After calculation and normalization, the stress index was constructed.

[0053] According to the spectral index calculation formula, calculate the values ​​of each band in the hyperspectral data, and combine the formula to obtain the spectral index value of the sample;

[0054] The spectral index value was used as the feature input, and the stress classification of the samples was performed according to the stress index using the random forest classification model.

[0055] Another object of the present invention is to provide a rice spectral stress index construction system, which aims to solve the technical problems of rapid and accurate identification of rice stress conditions, including the deficiencies of traditional methods in monitoring efficiency, accuracy and real-time performance, and the complexity of hyperspectral image data processing and analysis. By integrating multiple modules such as leaf photography, image processing, feature band extraction and model classification, the system can simplify the data processing process, improve analysis efficiency, reduce the uncertainty of stress cause analysis, and ultimately achieve real-time and automation of rice stress monitoring, providing precise management guidance for agricultural production.

[0056] To solve the above technical problems, the present invention provides the following technical solutions: a rice spectral stress index construction system, comprising: a leaf shooting module, which uses a portable hyperspectral camera to shoot rice leaves of different groups to obtain original hyperspectral images; an image processing module, which performs black and white correction on the original hyperspectral image to obtain a corrected image, extracts a region of interest and calculates its average spectral data, and performs SG smoothing filter processing on the average spectral data; a feature band extraction module, which uses a generative adversarial network to extract 50 feature bands with the highest contribution, and screens the feature bands according to the correlation with drought, fertilizer damage, and disease stress; a model classification module, which inputs the average spectral data corresponding to the feature band as model input data into a random forest classification model to obtain a model classification result; a stress classification module, which constructs a stress index according to the weight value of the feature band, divides the stress category, and outputs the final stress cause analysis.

[0057] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and wherein when the processor executes the computer program, the steps of the method described in the configuration of the method for constructing a rice spectral stress index are implemented.

[0058] A computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of the method described in the configuration of the method for constructing a rice spectral stress index are implemented.

[0059] Beneficial effects of the present invention: The present invention has the ability to accurately extract characteristic bands, and extracts characteristic bands from hyperspectral data based on a generative adversarial network. Compared with traditional methods (such as PCA and LLE), it can more efficiently capture nonlinear relationships in the data, extract characteristic bands with higher discrimination and diagnostic significance, and improve the performance of the classification model, providing more accurate data support for feature extraction and classification of rice hyperspectral images under different stresses. It has high computational efficiency and strong adaptability. Through the extraction of characteristic bands and the construction of stress indexes, the dimension of hyperspectral data is greatly reduced, and the calculation complexity of the model is effectively reduced. At the same time, the stress classification method based on spectral index can quickly and conveniently identify the health status and stress type of rice while maintaining accuracy. It has extensiveness and applicability, and meets the needs of various stress monitoring and analysis in rice planting management. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work, among which:

[0061] Figure 1 The overall flow chart of the method for constructing the rice spectral stress index provided by one embodiment of the present invention.

[0062] Figure 2 A generative adversarial network framework diagram of a method for constructing a rice spectral stress index provided by an embodiment of the present invention.

[0063] Figure 3 A diagram of the generative adversarial network structure of a method for constructing a rice spectral stress index provided by one embodiment of the present invention.

[0064] Figure 4 Spectral curves of rice leaves under different stresses in a method for constructing a rice spectral stress index provided by an embodiment of the present invention.

[0065] Figure 5 The invention provides a method for constructing a rice spectral stress index by visualizing the weights of spectral characteristic bands.

[0066] Figure 6 A confusion matrix diagram of classification results of a method for constructing a rice spectral stress index provided by an embodiment of the present invention.

[0067] Figure 7 The stress index classification result of the rice spectral stress index construction method provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0068] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0069] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0070] Example 1, reference Figure 1 , which is the first embodiment of the present invention, provides a method for constructing a rice spectral stress index, comprising:

[0071] Aiming at the task of rice stress factor analysis, the present invention relates to a method for extracting rice disease characteristic bands and constructing a stress index based on a generative adversarial network, accurately screening the sensitive bands of rice under different stress factors, and constructing a stress index based on the extracted sensitive bands. The method flow is as follows: Figure 1 shown.

[0072] Step S1: Use a portable hyperspectral camera to regularly photograph rice leaves of different groups on site to obtain original hyperspectral images.

[0073] Furthermore, based on the needs of extracting characteristic bands of rice and constructing stress index under different stresses, a portable hyperspectral camera is used to obtain real and effective spectral information of rice leaves through live shooting.

[0074] The portable hyperspectral camera is an internal push-scan device with a spectral wavelength range of 400 to 1200 nm, a spectral resolution of 2.5 nm, a spatial resolution of 1920×1920, and 1200 spectral channels.

[0075] Data collection was carried out in a field environment. Four groups of materials, namely healthy, drought-stressed, fertilizer-stressed, and disease-stressed, were set up in the rice experimental field as experimental objects. The hyperspectral image data of rice leaves under different stresses were recorded through regular photography, and the parts of rice leaves with obvious disease responses in the group were selected as typical samples.

[0076] All rice groups were grown in regularly monitored experimental fields to ensure that environmental factors were consistent, with only stress factors as variables.

[0077] Specifically, during the shooting process, natural sunlight is used as the light source. According to the changes in light intensity, the light signal amount of the hyperspectral camera is dynamically adjusted to ensure that the signal amount remains between 3000-4000 to obtain the best spectral data quality.

[0078] When shooting, the lens focuses precisely on the area where the rice leaves show symptoms, maximizing the capture of the symptoms under the stress, while avoiding interference from other non-stress-affected areas as much as possible. At the same time, in order to improve data consistency and reliability, the experimental site selects an appropriate shooting distance and angle to ensure that the sample is evenly illuminated.

[0079] A white plate with a fixed reflectivity of 0.9 is used as a reference to reduce background interference and improve the accuracy of subsequent data correction.

[0080] The resolution of the captured hyperspectral image is 1920×1920, and the shooting parameters have been repeatedly debugged, including exposure time and focus distance, to ensure the clarity and high signal-to-noise ratio of the hyperspectral image and avoid image distortion caused by lighting changes or equipment movement.

[0081] Through the internal push-scan acquisition mode of the portable hyperspectral camera equipment, regular and multi-period hyperspectral images of rice samples are collected to provide high-quality field data support for subsequent band sensitivity analysis feature extraction and stress index construction.

[0082] Step S2: Perform black-and-white correction on the original hyperspectral image to obtain a corrected hyperspectral image.

[0083] Specifically, due to the existence of dark current in the hyperspectral camera and the uneven intensity distribution of the light source under different spectra, the obtained hyperspectral image will be unstable.

[0084] Therefore, it is necessary to use the black and white correction method to correct the hyperspectral image to obtain a usable hyperspectral image.

[0085] Calculate the image reflectance value after black and white correction : , in, is the reflectance value of the original image to be corrected, is the reflectivity value obtained by scanning the standard reference white plate, is the blackboard reflectivity value obtained after the lens cover is put on.

[0086] After black and white correction, the corrected hyperspectral image data were further processed using ENVI software.

[0087] Step S3: extracting regions of interest from rice leaves under different stresses and calculating average spectral data.

[0088] Furthermore, the ENVI software was used to screen the areas with the most significant correlation with the stress factors from the corrected hyperspectral image data. When marking the areas, the areas with significant symptoms were defined as the areas of interest;

[0089] When rice leaves were under drought stress, fertilizer stress, or disease stress, the region of interest size was set to 50 × 50 pixels, and the average spectral data of each region was extracted;

[0090] For each leaf, n regions of interest were marked according to the shooting period and the stress symptoms actually observed, and the average spectral data records of rice leaves under n specific stresses were obtained.

[0091] When the rice leaves are healthy leaves, the regions of interest are randomly sampled using the five-point method, and n regions of interest are also marked, and the corresponding average spectral data are extracted.

[0092] Step S4: Perform SG smoothing filter processing on the average spectrum data.

[0093] Specifically, the data of all hyperspectral images are collected at the pixel level.

[0094] In order to reduce the interference of random noise in spectral data on analysis, the spectral data of each region of interest was processed with a Savitzky-Golay (SG) smoothing filter to eliminate random noise in the spectrum signal and improve the signal-to-noise ratio of the spectral signal.

[0095] The spectral data of each band corresponding to each pixel in the region of interest are averaged to obtain the average spectral representation of the region.

[0096] The average spectral data of all ROIs under different stress factors were summarized to form a spectral data set, which provided a basis for characteristic band extraction and stress index construction.

[0097] Step S5: Use the generative adversarial network to extract the 50 most contributing feature bands.

[0098] It should be noted that the generative adversarial network was used to extract the characteristic bands of rice leaves under different stresses.

[0099] The generative adversarial network consists of a generator and a discriminator. Through adversarial training, it can effectively capture the deep features in the data and is suitable for extracting band features in spectral images.

[0100] Compared with traditional feature extraction methods, generative adversarial networks can automatically learn features based on potential patterns in the data without relying on predefined feature screening methods. Their adaptive ability makes them more flexible in band selection and can extract spectral features of rice leaves under different stresses. At the same time, the generator can effectively learn the complex distribution of disease bands during the adversarial process, providing strong support for identifying characteristic bands of different stress categories.

[0101] In the generative adversarial network, the convolution layer of the discriminator has fully learned the local and global features of the real spectral data in the process of distinguishing true and false data. The feature weights of each band are extracted from the convolution layer of the discriminator, and sensitive feature bands are selected based on them.

[0102] Specifically, the framework structure of the generated adversarial network is as follows Figure 2 , Figure 3 As shown in Figure 1, the model consists of a generator and a discriminator, and combines the convolutional block attention mechanism to improve feature extraction and generation effects.

[0103] The generator includes a fully connected layer, a batch normalization layer, a random dropout layer, and a CBAM module to enhance the stability of training and the focusing ability of attention.

[0104] The CBAM module includes channel attention and spatial attention;

[0105] The channel attention includes extracting the channel information features of the input feature map through global average pooling and global maximum pooling operations, inputting the extracted features into a shared two-layer fully connected network, generating channel attention weights through addition operations after activation, and multiplying the input feature map channel by channel to complete weighted processing;

[0106] The spatial attention includes performing average pooling and maximum pooling operations on the weighted feature map in the channel attention dimension, compressing the channel dimension to 1, concatenating the two features, extracting spatial features through 7×7 convolution, and generating spatial attention weights after activation, which are multiplied element by element with the channel weighted feature map to complete the final weighting;

[0107] The CBAM module combines channel and spatial attention in a serial manner, which can adaptively refine the input feature map while maintaining the advantages of low computational overhead and easy integration.

[0108] The initial input of the generator is part of the real data and a random noise vector, which is expanded to 512 dimensions through the first layer of fully connected network, and then undergoes batch normalization and random inactivation with a ratio of 0.3. It then enters the second layer of fully connected network to expand the features to 1024 dimensions, and then undergoes batch normalization and random inactivation again. It then enters the third layer of fully connected network to further expand the features to 2048 dimensions, and then undergoes batch normalization and random inactivation again.

[0109] The features are reshaped into a three-dimensional tensor of (16, 16, 8), and the spatial attention and channel attention of the features are optimized through the CBAM module. The features are converted into one-dimensional vectors through the Flatten layer, and a 306-dimensional output vector is generated through the fully connected layer, representing the reflectance band value of the target hyperspectral data;

[0110] The discriminator includes using a 3D convolutional network combined with an attention mechanism to judge the input data; reshaping the input data into a three-dimensional data block, and extracting three-dimensional spatial features through two sets of 3D convolutional layers and a 3D maximum pooling layer.

[0111] The first set of 3D convolutional layers contains 32 filters, and the second set contains 64 filters, both with kernel size (3, 3,3).

[0112] The CBAM module is used to enhance the perception of feature weights. The data is further extracted through two groups of 2D convolutional layers and 2D maximum pooling layers (each group contains 32 and 64 filters respectively, with a kernel size of (3, 3).

[0113] The CBAM module is introduced again after the second set of convolutions to strengthen the feature attention, and an inactivation layer is added to improve the generalization ability of the model.

[0114] The data passes through the flattening layer and is passed to the fully connected layer for feature integration, and the discrimination result is output.

[0115] Among them, both the generator and the discriminator focus on key features through the CBAM module. The generator pays attention to reconstructing the details of the target data, while the discriminator enhances the classification and discrimination capabilities with multi-dimensional convolutional features. The overall model can effectively handle complex hyperspectral data tasks.

[0116] The generative adversarial network model is based on the PyTorch deep learning framework, version 2.2.1. In terms of GPU acceleration, CUDA 11.1 and cuDNN 8.9.6 are used to improve the efficiency of model training and reasoning.

[0117] The discriminator is trained together with the generator to optimize the model performance in an adversarial manner.

[0118] During model training, the batch size is set to 32, half of the batch is used for real data, and the other half is used for fake data generated by the generator.

[0119] The discriminator is trained based on real data and generated data respectively, and the maximum training loss of the two is calculated, with a maximum number of iterations of 10,000.

[0120] The Adam optimizer is used, the initial learning rate is set to 0.001, and the weights of the generator and discriminator are updated simultaneously.

[0121] The loss function uses binary cross entropy to evaluate the adversarial effect between the generator and the discriminator, and monitors the convergence and adversarial balance of the model in real time.

[0122] To avoid overfitting and improve the generalization ability of the generator, a model checkpoint mechanism is added to the model training to save only the model parameters with the best generation quality. At the same time, a validation set of training data is introduced to monitor the performance of the generator on real data to ensure that the model converges to the optimal solution.

[0123] It should also be noted that by analyzing the activation maps of the discriminator at different convolutional layers and combining the gradient of the convolutional layer weights, the sensitivity of the convolution kernel to different bands is located, and the selected feature bands and corresponding weights are extracted; this method can directly reveal the contribution of each band in the extraction of spectral features, making it easier to screen out bands with greater diagnostic value.

[0124] After obtaining the feature weights in the convolution layer, the bands with high weight values ​​are screened out, and then the sensitive bands with the strongest correlation with drought stress, fertilizer stress, and disease stress are screened again;

[0125] By sorting and screening the weights of the sensitive feature bands output by the convolutional layer, the top 50 bands with the most contributions from 400nm to 1200nm are determined, and the spectral reflectance data corresponding to the current 50 bands are retained as the input of the model.

[0126] Optimize the input features of the model to improve the accuracy of data classification under different stress groups of rice.

[0127] Figure 4 , Figure 5 The characteristic curves of rice leaf spectral reflectance under different types of stress and the visualization diagrams of the screened spectral characteristic band weights are shown respectively.

[0128] Step S6: input the average spectral data corresponding to the characteristic band into the random forest classification model as model input data to obtain the model classification result.

[0129] It should be noted that the classification model uses the classic Random Forest (RF) model to classify the data set. The RF model is an ensemble learning method and an extension of the decision tree model. It achieves classification or regression by building multiple decision trees and combining voting or averaging methods. The RF model has good anti-overfitting ability and efficient classification performance, and is particularly suitable for high-dimensional data and feature-selected data sets.

[0130] Step S7: Simultaneously select the top 5 characteristic bands and the corresponding weights to construct a spectral index, and obtain the stress index to classify the stress categories.

[0131] Specifically, the dataset after characteristic band screening was used as the input of the random forest model, and the average spectral dataset was labeled according to different stress groups, including 0 for health, 1 for drought stress, 2 for fertilizer stress, and 3 for disease stress;

[0132] The random forest model includes a training phase and a prediction phase;

[0133] The training phase includes random forests using the bagging method to randomly sample from the training data set to generate m data subsets containing some samples; building a decision tree on each subset, and randomly selecting a subset of features for splitting during the node division process of each tree.

[0134] The prediction stage includes random forest using all constructed decision trees to predict the test samples and determine the final category result through majority voting method;

[0135] The output of the random forest model is represented as: , in, For the final prediction result, is the total number of decision trees in the random forest, For the The prediction results of each tree;

[0136] The output of the random forest model is the corresponding stress group.

[0137] It should be noted that in order to verify the superiority of the generative adversarial network in feature band extraction, the present invention demonstrates the classification effect of this method after feature extraction by comparing it with principal component analysis and local linear embedding methods.

[0138] This method is based on generative adversarial networks for feature extraction, which can better learn the potential structure and nonlinear relationships in the data. Through the adversarial training mechanism, the model can automatically optimize and generate more discriminative features during the feature extraction process, thereby improving the performance of the classification model. Compared with PCA and LLE, the features extracted by this method show better classification results on complex hyperspectral data.

[0139] After obtaining the weights of the bands, the five bands with the most active response of rice leaves under the stress of this experiment were screened and sorted according to the weight values. The corresponding coefficients of the bands were obtained from the weights. After calculation and normalization, the stress index k was constructed: , in, , , , , Respectively represent the reflectance values ​​of the screening bands, each band corresponds to a weight coefficient, and min and max represent Computed global minimum and maximum values.

[0140] Furthermore, the stress index is applied to classify the samples according to their stress.

[0141] According to the selected spectral index calculation formula, the values ​​of each band in the hyperspectral data are calculated, and the spectral index value of the sample is obtained by combining the formula; for example, Represents the spectral reflectance value corresponding to the wavelength of 572 nm. Each data sample contains the reflectance value corresponding to the effective range of 400nm-1000nm. Take the reflectance value under nm corresponding to the stress index k and substitute it into the formula to calculate the stress index value k corresponding to the sample.

[0142] These spectral index values ​​can effectively reflect the spectral characteristics of rice under different stress conditions, especially the changes related to stresses such as drought, fertilizer damage, and disease.

[0143] The calculated index value is used as the feature input, and the stress index is constructed based on the feature bands and weights selected by the generative adversarial network. The stress index is used as the new data input and input into the random forest model for stress classification: 0 represents health, 1 represents drought stress, 2 represents fertilizer stress, and 3 represents disease stress.

[0144] Despite the use of a single spectral index value, the model can still accurately distinguish different stress categories in hyperspectral data, demonstrating the effectiveness of spectral indices in stress classification. In addition, the features extracted using spectral indices greatly reduce the dimensions of the data and the amount of computation required for the model, thereby improving classification efficiency and computing speed. This method not only optimizes computing resources, but also provides strong support for large-scale plant stress monitoring and precision agriculture applications, and has broad practical application prospects.

[0145] Step S8: The model classification result and the stress category are used to determine the group to which the leaf belongs. The model classification result directly outputs the label corresponding to the sample. According to the category corresponding to the label, the group corresponding to the sample is determined, and the cause of rice stress is output.

[0146] In summary, this patent proposes a method for extracting characteristic bands of rice diseases and constructing a stress index based on generative adversarial networks and hyperspectral images, which effectively improves the feature extraction and classification effects of hyperspectral image data in rice stress monitoring. Through this method, characteristic bands related to rice stress can be extracted more accurately, and stress classification can be performed based on spectral indices, which improves classification accuracy and efficiency. It not only reduces the data dimension and model calculation amount, but also provides rice growers with more scientific and effective stress monitoring and decision-making support, which helps to improve the management level of rice production and promote sustainable development in the agricultural field.

[0147] Example 2 is an embodiment of the present invention, which provides a rice spectral stress index construction system, including: a leaf shooting module 100, an image processing module 200, a feature band extraction module 300, a model classification module 400, and a stress classification module 500.

[0148] The leaf shooting module 100 uses a portable hyperspectral camera to shoot rice leaves of different groups to obtain original hyperspectral images.

[0149] The image processing module 200 performs black-and-white correction on the original hyperspectral image to obtain a corrected image, extracts the region of interest and calculates its average spectral data, and performs SG smoothing filter processing on the average spectral data.

[0150] The characteristic band extraction module 300 uses a generative adversarial network to extract the 50 characteristic bands with the highest contribution, and selects the characteristic bands according to their correlation with drought, fertilizer damage, and disease stress.

[0151] The model classification module 400 inputs the average spectral data corresponding to the characteristic band as model input data into the random forest classification model to obtain a model classification result.

[0152] The stress classification module 500 constructs a stress index according to the weight values ​​of the characteristic bands, classifies the stress categories, and outputs the final stress cause analysis.

[0153] Embodiment 3, the third embodiment of the present invention, is different from the first two embodiments in that:

[0154] Specifically, if the rice spectral stress index construction method is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or part of the technical solution that contributes to the prior art or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including 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 method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0155] The computer program includes several instructions that can cause a computer device (such as a server, a personal computer or other network device) to perform the following steps:

[0156] A portable hyperspectral camera was used to regularly photograph rice leaves of different groups on site to obtain original hyperspectral images; the original hyperspectral images were corrected to black and white to obtain corrected hyperspectral images; the regions of interest were extracted from rice leaves under different stresses, and the average spectral data were calculated; the average spectral data was processed with an SG smoothing filter; the 50 most contributing feature bands were extracted using a generative adversarial network; the average spectral data corresponding to the feature bands were input into a random forest classification model as model input data to obtain the model classification results; at the same time, the top 5 feature bands and corresponding weights were screened to construct a spectral index, and the stress index was obtained to divide the stress categories; the model classification results and stress categories were used to determine the group to which the leaves belonged, and the cause of rice stress was output.

[0157] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0158] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0159] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0160] The computer program of the present invention further supports real-time operation. For example, the present invention uses a generative adversarial network to extract sensitive bands, which can quickly and automatically learn and identify features in hyperspectral data without manually selecting features. This intelligent feature extraction method significantly improves the processing speed, enables the model to be updated and optimized in real time, and is suitable for changing field environments. Each step involved in the present invention (such as black and white correction, region of interest extraction, spectral data smoothing, etc.) is automatically processed using computer software (such as ENVI software), which reduces the time of manual intervention and improves processing efficiency. This automated process ensures that the required spectral data can be quickly acquired and data analysis can be performed in a timely manner.

[0161] Example 4, reference Figure 6-Figure 7 , which is an embodiment of the present invention, provides a method for constructing a rice spectral stress index. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through experiments.

[0162] Figure 6 The classification performance of this method in four categories, namely health, drought stress, fertilizer stress, and disease stress, is shown. From the confusion matrix, we can see:

[0163] The number of samples in the category “0-healthy” that were correctly classified as healthy was 228, with no misclassification;

[0164] The number of samples correctly classified as category “1-drought” was 248, and only a small number of samples (4) were misclassified as “2-fertilizer damage”;

[0165] The category “2-fertilizer damage” was correctly classified as 236, 8 samples were misclassified as “0-healthy”, and 4 were misclassified as “3-disease”;

[0166] The category “3-disease” was correctly classified as 228, and 4 samples were misclassified as “2-fertilizer damage.

[0167] After using the generative adversarial network to extract feature bands, the random forest classification model achieved a classification accuracy of 97.92%, and the classification results were visualized using a confusion matrix. After feature extraction, a large amount of redundant data can be eliminated, greatly reducing data input, and still achieving a high classification accuracy, especially for the "healthy", "drought" and "disease" categories, with almost no misclassification, reflecting the strong ability of this feature extraction method to distinguish different stress states of rice.

[0168] As shown in Table 1, the features extracted by this method perform better in multiple evaluation indicators such as classification accuracy, average accuracy, and Kappa coefficient.

[0169] Specifically, Table 1 shows the classification performance comparison between the feature extraction method and traditional feature extraction methods such as PCA and LLE under the random forest classifier. It can be seen that the method is superior to the traditional PCA and LLE feature extraction methods in all indicators. This shows that the feature extraction method of the present invention can more accurately capture the hyperspectral features of rice under different stress conditions and improve the classification effect. Table 1 is shown below: Table 1 ,

[0170] Based on the spectral index value as a single feature input, the classification model achieved an accuracy of 85%. The classification results can be visualized through the confusion matrix, such as Figure 7 As shown, the classification status and error distribution between categories are further demonstrated.

[0171] Specifically, Figure 7 The stress index value is used for classification. Compared with the full band and the first 50 feature bands extracted by this method (each data has 50 feature values), it is further used to input only a single feature value (stress index value) into the random forest classification model, and a classification accuracy of 85% is obtained.

[0172] Using a single stress index value to further simplify feature input, although the classification performance has decreased, it still reaches 85%, indicating that the stress index is highly representative; compared with traditional methods, the stress index designed by the present invention can still maintain a high classification ability under the condition of low-dimensional features, indicating that it has strong application potential in actual agricultural monitoring; in hyperspectral image analysis, using full-band features or the first 50 bands has better classification performance, but the computational complexity is higher, and more resources are required for model training and prediction. A single stress index value greatly reduces the data dimension, is lightweight, and is suitable for resource-constrained scenarios, such as mobile devices or real-time monitoring systems.

[0173] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for constructing a rice spectral stress index, characterized in that: include, A portable hyperspectral camera was used to regularly photograph rice leaves of different groups on site to obtain original hyperspectral images. Perform black and white correction on the original hyperspectral image to obtain a corrected hyperspectral image; The regions of interest were extracted from rice leaves under different stresses, and the average spectral data were calculated; The average spectral data were processed with SG smoothing filter; Use the generative adversarial network to extract the 50 most contributing feature bands; The average spectral data corresponding to the characteristic band is input into the random forest classification model as the model input data to obtain the model classification result; At the same time, the top 5 characteristic bands and corresponding weights were selected to construct the spectral index, and the stress index was obtained to divide the stress categories; The model classification results and stress categories are used to determine the group to which the leaf belongs, and the cause of rice stress is output.

2. The method for constructing a rice spectral stress index according to claim 1, characterized in that: The original hyperspectral image, include, A portable hyperspectral camera was used, natural sunlight was used as the light source, and a white board with a fixed reflectivity of 0.9 was used as a reference; Obtain spectral information of rice leaves through live photography; Dynamically adjust the light signal of the hyperspectral camera to maintain at 3000-4000; Data collection was conducted in a field environment, with regular photography of healthy, drought-stressed, fertilizer-stressed, and disease-stressed rice leaves in the rice experimental field; Recording hyperspectral image data of rice leaves under different stresses; The rice leaf parts with obvious disease response in the group were selected as typical samples; The resolution of the captured hyperspectral image is 1920×1920.

3. The method for constructing a rice spectral stress index according to claim 1 or 2, characterized in that: The black and white correction includes: The hyperspectral image is corrected by using the black-and-white correction method to obtain a corrected hyperspectral image; Calculate the image reflectance value after black and white correction : in, is the reflectance value of the original image to be corrected, is the reflectivity value obtained by scanning the standard reference white plate, is the blackboard reflectivity value obtained after the lens cover is put on.

4. The method for constructing a rice spectral stress index according to claim 3, characterized in that: The region of interest extraction includes: ENVI software was used to screen the areas with the most significant correlation with stress factors from the corrected hyperspectral image data and define them as areas of interest; When rice leaves were under drought stress, fertilizer stress, or disease stress, the region of interest size was set to 50 × 50 pixels, and the average spectral data of each region was extracted; According to the shooting period and the actually observed stress symptoms, n ROIs were marked to obtain the average spectral data records of rice leaves under n specific stresses; When the rice leaves were healthy, the region of interest was randomly sampled using the five-point method, n ROIs were also marked, and the corresponding average spectral data were extracted.

5. The method for constructing a rice spectral stress index according to claim 2 or 4, characterized in that: The generative adversarial network comprises: Generator, Discriminator; The generator includes a fully connected layer, a batch normalization layer, a random inactivation layer and a CBAM module; The initial input of the generator is part of the real data and a random noise vector, which is expanded to 512 dimensions through the first layer of fully connected network, and then undergoes batch normalization and random dropout with a ratio of 0.

3. The second layer of fully connected network expands the features to 1024 dimensions, and then undergoes batch normalization and random dropout again. The third layer of fully connected network further expands the features to 2048 dimensions, and then undergoes batch normalization and random dropout again. The features are reshaped into a three-dimensional tensor of (16, 16, 8), and the spatial attention and channel attention of the features are optimized through the CBAM module. The features are converted into one-dimensional vectors through the Flatten layer, and a 306-dimensional output vector is generated through the fully connected layer, representing the reflectance band value of the target hyperspectral data; The discriminator includes using a 3D convolutional network combined with an attention mechanism to judge the input data; The input data is reshaped into a three-dimensional data block, and three-dimensional spatial features are extracted through two sets of 3D convolutional layers and 3D maximum pooling layers; The first set of 3D convolutional layers contains 32 filters, and the second set of 3D convolutional layers contains 64 filters, and the kernel size is (3,3,3); After the CBAM module enhances the perception of feature weights, the data passes through two sets of 2D convolutional layers and 2D maximum pooling layers. After the second set of convolutions, the CBAM module is introduced again to strengthen the attention of the features, and an inactivation layer is added. The features are integrated through the flattening layer and passed to the fully connected layer to output the discrimination results.

6. The method for constructing a rice spectral stress index according to claim 5, characterized in that: The 50 characteristic bands with the highest contribution to extraction include: Analyze the activation map of the discriminator at different convolutional layers, combine the gradient of the convolutional layer weights, locate the sensitivity of the convolution kernel to different bands, and extract the selected feature bands and corresponding weights; After obtaining the feature weights in the convolution layer, the bands with high weight values ​​are screened out, and then the sensitive bands with the strongest correlation with drought stress, fertilizer stress, and disease stress are screened again; By sorting and screening the weights of the sensitive feature bands output by the convolutional layer, the top 50 bands with the most contributions from 400nm to 1200nm are determined, and the spectral reflectance data corresponding to the current 50 bands are retained as the input of the model.

7. The method for constructing a rice spectral stress index according to claim 1, 2, 4 or 6, characterized in that: The model classification results include: The dataset after characteristic band screening was used as the input of the random forest model, and the average spectral dataset was labeled according to different stress groups, including 0 for health, 1 for drought stress, 2 for fertilizer stress, and 3 for disease stress; The random forest model includes a training phase and a prediction phase; The training phase includes random forest randomly sampling from the training data set using the bagging method to generate m data subsets containing partial samples; Build a decision tree on each subset, and randomly select a subset of features to split during the node partitioning process of each tree; The prediction stage includes using all constructed decision trees to predict the test samples using random forests, and determining the final category result by majority voting; The output of the random forest model is represented as: in, For the final prediction result, is the total number of decision trees in the random forest, For the The prediction results of each tree; The output of the random forest model is the corresponding stress group.

8. The method for constructing a rice spectral stress index according to claim 1, 2, 4 or 6, characterized in that: The classification of coercion includes: The five bands with the most positive response of rice leaves under stress were selected and sorted according to the weight values ​​of the characteristic bands, and the corresponding coefficients of the bands were obtained from the weights. After calculation and normalization, the stress index was constructed. According to the spectral index calculation formula, calculate the values ​​of each band in the hyperspectral data, and combine the formula to obtain the spectral index value of the sample; The spectral index value was used as the feature input, and the stress classification of the samples was performed according to the stress index using the random forest classification model.

9. A rice spectral stress index construction system, characterized in that: include, The leaf shooting module (100) uses a portable hyperspectral camera to shoot rice leaves of different groups to obtain original hyperspectral images; An image processing module (200) performs black-and-white correction on the original hyperspectral image to obtain a corrected image, extracts a region of interest and calculates its average spectral data, and performs SG smoothing filter processing on the average spectral data; A feature band extraction module (300) uses a generative adversarial network to extract 50 feature bands with the highest contribution, and selects the feature bands according to their correlation with drought, fertilizer damage, and disease stress; A model classification module (400) inputs the average spectral data corresponding to the characteristic band as model input data into a random forest classification model to obtain a model classification result; The stress classification module (500) constructs a stress index according to the weight value of the characteristic band, classifies the stress categories, and outputs the final stress cause analysis.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method for constructing the rice spectral stress index are implemented.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for constructing a rice spectral stress index are implemented.

Images