Detection Model for Asymptomatic Stage of Cotton Verticillium Wilt Based on Sensitive Wavelet Features

Abstract

The present invention discloses a detection model for asymptomatic stage of cotton Verticillium wilt constructed based on sensitive wavelet features. The construction method of the detection model for asymptomatic stage of cotton Verticillium wilt constructed based on sensitive wavelet features is as follows: first, inoculate cotton with Verticillium wilt and confirm that the cotton is infected with Verticillium wilt; then obtain the spectral information of cotton leaves at different inoculation dates; preprocess the spectral data through continuous wavelet transform; use a scanning electron microscope to obtain the internal structure feature data of healthy and asymptomatic Verticillium wilt-infected cotton leaves; based on the internal structure feature data of cotton leaves, and use LASSO to estimate the regression coefficient β j by minimizing the objective function, so as to select the optimal sensitive wavelet features; finally, through the selected sensitive wavelet feature set, use the support vector machine algorithm to determine the threshold for distinguishing healthy and Verticillium wilt-infected leaves by fusing the wavelet feature set, thereby realizing the detection of the asymptomatic stage of cotton Verticillium wilt.

Description

Technical Field

[0001] The present invention relates to a method for detecting asymptomatic stage of cotton Verticillium wilt and a method for constructing a detection model based on hyperspectral sensitive wavelet information. Background Art

[0002] Cotton is an important fiber crop and oil crop in the world. Verticillium wilt (VW) is one of the most common and destructive diseases during the growth and development of cotton. It spreads pathogens through the soil, enters the host from the root system, multiplies in the vascular tissue of the plant, and causes Verticillium wilt. The prevention and control of Verticillium wilt are very difficult. According to statistics, the average annual global cotton output is 23.52 million tons, and the annual yield loss caused by Verticillium wilt is as high as 30%. How to detect and control the spread of Verticillium wilt disease early has become the current research difficulty and focus.

[0003] The pathogenic mechanism of VW is to secrete "toxins" to degrade the root cell wall and occupy the xylem of the plant. At this stage, Verticillium dahliae will multiply in the xylem for 20 - 30 days, or even lie dormant for a longer time, before obvious symptoms can be seen on the plant surface. When obvious symptoms appear, the Verticillium wilt pathogen has already colonized the plant in large numbers, causing irreversible damage and seriously affecting the yield and quality of the crop. Therefore, detecting that cotton is under the stress of Verticillium wilt when the disease occupies the xylem of the plant and before symptoms appear (asymptomatic stage) will be beneficial to prevent and control the occurrence and development of Verticillium wilt. At present, the traditional detection method for cotton Verticillium wilt is still based on field visual inspection, and the disease condition is judged by the "external" symptoms of cotton Verticillium wilt. This method is difficult to diagnose effectively. Once detected, it may already be in the "mid - late stage", causing significant losses. At the same time, the detection technology based on indoor fungal culture by field sampling, although it can achieve accurate early diagnosis, requires destructive sampling, is time - consuming and laborious, and is expensive, and cannot meet the needs of large - scale early and rapid diagnosis of cotton VW in actual production. In order to quickly and non - destructively detect Verticillium wilt, a more effective method is needed.

[0004] With the continuous development of remote sensing monitoring technology, the detection accuracy before the symptoms of leaf pathogens infection has been greatly improved. Some studies have used spectral technology to accurately detect the physiological characteristic changes in the early symptomatic stage of Verticillium wilt, so as to achieve the accurate monitoring of Verticillium wilt. However, the current early monitoring technology for plant infection with Verticillium wilt still conducts accurate monitoring after symptoms appear on plant leaves. There are few studies exploring the variation laws of the spectra and chlorophyll fluorescence of the main stem leaves of plants in the asymptomatic stage of Verticillium wilt infection, as well as using these technologies for the accurate diagnosis of Verticillium wilt in the asymptomatic stage. In the asymptomatic stage of Verticillium wilt-infected plants, effector factors secreted by Verticillium dahliae cause host immune responses, disrupt the plant hormone balance, thereby promoting the development of Verticillium wilt, leading to plant vascular blockage and oxidative stress responses. The connection between such physiological symptoms and defense mechanisms, which results in the changes in leaf structure and substances, will provide a theoretical basis for the spectral detection of Verticillium wilt in the asymptomatic stage. In the asymptomatic infection stage of pathogens, the subtle changes in the internal structure of leaves and the changes in photosynthetic capacity are difficult to be characterized by the original spectral features, and some subtle characteristic changes on leaves are difficult to capture. Continuous wavelet transform (CWT), as a promising weak feature extraction tool, can capture subtle spectral absorption features and enhance the subtle spectral signals caused by pathogen infection. Therefore, based on the analysis of the above problems, the present invention provides a detection method and a detection model construction method for the asymptomatic stage of cotton Verticillium wilt based on hyperspectral sensitive wavelet information. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention provides a detection method and a detection model construction method for the asymptomatic stage of cotton Verticillium wilt based on hyperspectral sensitive wavelet information, so as to solve the problem that it is difficult to detect cotton infected with Verticillium wilt in time in the asymptomatic stage.

[0006] The present invention is implemented by the following technical solutions:

[0007] A detection model for the asymptomatic stage of cotton Verticillium wilt constructed based on sensitive wavelet features, and the construction method of the detection model is as follows:

[0008] S1. Inoculate cotton with Verticillium wilt and confirm that the cotton is infected with Verticillium wilt;

[0009] S2. Obtain the spectral information of cotton leaves at different inoculation dates;

[0010] S3. Preprocess the spectral data through continuous wavelet transform;

[0011] S4. Use a scanning electron microscope to obtain the internal structure characteristic data of healthy and asymptomatic Verticillium wilt-infected cotton leaves;

[0012] S5. Based on the internal structure characteristic data of cotton leaves, and using LASSO to estimate the regression coefficient β by minimizing the objective function, j so as to select the optimal sensitive wavelet features.

[0013]

[0014] where y is the leaf type, and x is the normalized reflectance of the wavelength in the function; n and p are the number of samples and the wavelength respectively; β j is the coefficient, α is the intercept, and γ is the penalty term for controlling the shrinkage value;

[0015] S6. Construct a detection model for the asymptomatic stage of cotton Verticillium wilt. Through the sensitive wavelet feature set screened by S5, using three widely used machine learning methods, namely support vector machine (SVM), logistic regression algorithm, and K-nearest neighbor algorithm (KNN), the threshold for distinguishing healthy and Verticillium wilt-infected leaves is determined by fusing the wavelet feature set, so as to realize the detection of the asymptomatic stage of cotton Verticillium wilt.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] The present invention provides a detection model for the asymptomatic stage of cotton Verticillium wilt constructed based on sensitive wavelet features. Through the screened sensitive wavelet feature set, using the support vector machine algorithm, the threshold for distinguishing healthy and Verticillium wilt-infected leaves is determined by fusing the wavelet feature set, so as to realize the detection of the asymptomatic stage of cotton Verticillium wilt. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a flowchart for constructing a detection model for the asymptomatic stage of cotton Verticillium wilt in the present invention.

[0019] DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] The present invention will be further described below with reference to specific embodiments.

[0021] Embodiment 1

[0022] I. Preparation and inoculation method of cotton Verticillium wilt

[0023] Inoculate Verticillium dahliae (strain V952) into the sucrose sodium nitrate liquid medium, and incubate it under the conditions of 25°C and 200 rpm for 4 - 5 days. After filtering the colonies with sterilized gauze, use the plate counting spore method to dilute the spore concentration to 4×10 7 cfu·ml -1 . Then, pour the spore suspension into the cotton cultivation box in the form of irrigation, and rinse the control group with an equal amount of sterile water.

[0024] II. Determination method for the asymptomatic stage of Verticillium wilt

[0025] Cut the cotton stems and petioles inoculated with Verticillium dahliae into small sections of 0.5 - 0.8 cm, repeat twice for each part, use PDA medium, culture at 25 °C for 7 days, and take pictures to observe the fungal biomass. Cut the 2-cm stem segment above the cotyledon node, place it in a sterilized triangular flask, add 30 mL of sodium hypochlorite solution, and surface disinfect the stem segment for 5 minutes. After disinfection, rinse it 3 times with sterilized ddH2O. Place the rinsed stem segment in a petri dish lined with sterilized filter paper, cut it into 1-cm stem segments with a sterilized scalpel, and use sterilized forceps to neatly place the stem segments in the PDA medium, and observe whether Verticillium dahliae appears after culturing in the dark in a 25 °C constant temperature incubator for 7 days. II. Data Acquisition

[0026] 2.1 Acquisition of Microscopic Structure of Healthy and Diseased Leaves

[0027] Dehydrate the leaves after measuring the spectral information, and use a scanning electron microscope model SU8010 produced by Hitachi, Japan, to obtain the cross-sectional information of the leaves, and observe the structural changes of the palisade tissue and spongy tissue of the leaves after disease stress.

[0028] 2.2 Acquisition of Spectral Information of Healthy and Diseased Leaves

[0029] Use the SR-3500 portable ground object spectrometer developed by Spectral Evolution, USA, to measure the spectral data (350 - 2500 nm) of healthy cotton and cotton leaves under VW stress. During the measurement, use the built-in light source and leaf clip of the spectrometer for in vivo acquisition. When acquiring data, avoid the leaf veins of cotton leaves, measure 3 positions for each leaf, and take the average value of 5 repetitions at each position as the spectral value of this position. Calibrate the spectral data with the standard plate before and after each measurement.

[0030] III. Extraction of Characteristic Spectra of the Asymptomatic Period of Cotton Verticillium Wilt

[0031] Use the wavelet transform method to perform continuous scale transformation on the reflection spectrum to obtain the wavelet coefficient spectrum, and then extract the key components (wavelet features). This wavelet transform method can better express the spectral shape information, rather than just the intensity information of the spectral bands. The wavelet coefficients at multiple scales represent different meanings of the spectral information. The low-scale features correspond to high-frequency spectral changes (microscopic features), and the high-scale features correspond to low-frequency spectral changes (macroscopic features).

[0032] Its conversion formula is:

[0033]

[0034] Among them, a represents the wave width and b represents the phase. The spectral signal is decomposed by wavelet to obtain a complete energy coefficient matrix at different wavelengths and decomposition scales: In this study, the Mexican hat wavelet basis function is used. In this study, an arithmetic progression is adopted, and the CWT coefficients are set to 2, 4, 6, …, 14, 16, and a total of 8 wavelet dimension transformations are obtained, and different wavelet scales are represented by 1, 2, 3, …, 7, 8.

[0035] Through the results of analysis of variance (ANOVA), this study initially obtained that there were extremely significant sensitive wavelet features (p < 0.01) in the microscopic structure state of cotton leaves under healthy and VW-infected leaf conditions at different inoculation times under two environmental conditions. The regularization regression technique (LASSO) was used to estimate the value of the regression coefficient (βj) by minimizing the objective function. Finally, the variance inflation factor (VIF) was used to test the multicollinearity of sensitive wavelet features; wavelet features with low multicollinearity were screened out with VIF < 10 as the threshold.

[0036]

[0037] Among them, y is the leaf type, x is the normalized reflectance of the wavelength in the function; n and p are the number of samples and wavelengths respectively; β j is the coefficient, α is the intercept, and γ is the penalty term for controlling the shrinkage value. Generally speaking, γ is a non-negative regularization parameter corresponding to a value from 0 to 1. As γ increases, the number of non-zero components of β decreases, that is, the number of variables decreases. The 10-fold cross-validation is used to determine the coefficients of the regularized linear regression model. The absolute values of the coefficients are calculated and sorted to obtain the most important wavelet features.

[0038] Table 1 Sensitive wavelet features in the asymptomatic stage of cotton Verticillium wilt screened based on LASSO-VIF

[0039]

[0040] IV. Model construction A detection model for the asymptomatic stage of cotton Verticillium wilt based on spectral wavelet features was created using the extracted spectral wavelet features and the health conditions of cotton leaves, that is, accurate detection models for the asymptomatic stage of cotton Verticillium wilt were established using three common machine learning methods.

[0041] In this embodiment, three widely used machine learning methods, namely, Support Vector Machine (SVM), Logistic Regression Algorithm, and K-Nearest Neighbor Algorithm (KNN), were selected for the diagnosis of the asymptomatic period of cotton VW infection. SVM is a non-parametric supervised classifier that reduces the misclassification error of training data by minimizing the structural risk strategy. It was originally designed for two-class classification problems. By constructing an optimal separating hyperplane to maximize the margin between the classes with fewer support vectors (training samples), it is a soft classification strategy with strong generalization ability and good robustness (Cortes and Vapnik, 1995). The Logistic Regression Algorithm assumes that the data follows a Bernoulli distribution. By using the method of maximum likelihood function and applying gradient descent to solve the parameters, it aims to dichotomize the data. It is a simple classification model with strong model interpretability. KNN is a non-parametric classification method that classifies unlabeled samples by analyzing their K nearest neighbors. We used a grid search strategy to determine the model parameters C and γ, which are the expansion and regularization parameters of the RBF kernel, respectively. The classification algorithm was cross-validated 10 times and repeated 10 times to obtain the OA value and Kappa coefficient of the optimal classification accuracy.

[0042] V. Model Validation and Result Analysis

[0043] In this method, 40 pots of cotton were planted in different environments and inoculated with Verticillium dahliae. The selected wavelet features were used to test the detection accuracy of the asymptomatic period of cotton Verticillium wilt. Through ten-fold cross-validation, the precision rate and Kappa coefficient were selected as the accuracy detection indicators. The closer these two indicators are to 1, the higher the detection accuracy.

[0044] After verification, in the field scenario and greenhouse scenario, the detection accuracy of the models constructed under different inoculation dates is greater than 0.8, and the Kappa coefficient is greater than 0.6. It can accurately judge whether the asymptomatic leaves are infected with Verticillium wilt, which may provide a fast and convenient method for detecting the asymptomatic infection period of cotton VW in the field.

[0045] In summary, it can be seen that in this embodiment, the spectral wavelet features can be used to accurately detect the asymptomatic stage of cotton Verticillium wilt. Non-destructive and rapid detection of VW infection in the asymptomatic stage is of great significance for preventing and controlling the large-scale development of the disease and thus reducing the impact on yield. We explored the changes in the internal structure of cotton leaves in the asymptomatic stage of Verticillium wilt infection in different environmental scenarios over two years. At the same time, the wavelet features screened by VIF-LASSO in the two-year experiment showed high consistency, indicating the feasibility of hyperspectral wavelet signals in non-destructive detection of the asymptomatic stage of cotton VW. Through VIF-LASSO, the wavelet features we screened are all concentrated in the VNIR region, which is related to the physiological characteristics of internal structure damage and vessel blockage in the asymptomatic stage of the leaves we observed. At the same time, 4-5 wavelet features can be used to accurately identify cotton leaves infected asymptomatically, and their classification accuracy exceeds 80%, and the Kappa coefficient is higher than 0.6. Among them, the average accuracy of the detection model based on logistic regression analysis is as high as 90.62%.

[0046] It should be noted that the above are only several specific embodiments of the present invention. Obviously, the present invention is not limited to the above embodiments and there can be other variations. All variations directly derived or indirectly extended by those skilled in the art from the disclosed content of the present invention should be considered within the protection scope of the present invention.

Claims

1. A detection model for asymptomatic stage of cotton Verticillium wilt constructed based on sensitive wavelet features, characterized in that, The construction method of the detection model is as follows: S1. Inoculate cotton with Verticillium wilt and confirm that the cotton is infected with Verticillium wilt; S2. Obtain the spectral information of cotton leaves at different inoculation dates; S3. Preprocess the spectral data through continuous wavelet transform; S4. Use a scanning electron microscope to obtain the internal structural feature data of healthy cotton leaves and cotton leaves in the asymptomatic stage of Verticillium wilt infection; S5. Based on the internal structural feature data of cotton leaves and using LASSO to estimate the regression coefficient β by minimizing the objective function j to select the optimal sensitive wavelet features where y is the leaf type, and x is the normalized reflectance of the wavelength in the function; n and p are the number of samples and the wavelength, respectively; β j is the coefficient, α is the intercept, and γ is the penalty term for controlling the shrinkage value; S6. Construct a detection model for the asymptomatic stage of cotton Verticillium wilt. This model uses three widely used machine learning methods, namely support vector machine (SVM), logistic regression algorithm, and K-nearest neighbor algorithm (KNN), through the sensitive wavelet feature set screened in S5, and determines the threshold for distinguishing healthy and Verticillium wilt-infected leaves by fusing the wavelet feature set, so as to realize the detection of the asymptomatic stage of cotton Verticillium wilt.

Images