An early detection method for eggplant Verticillium wilt based on multispectral imaging technology

Abstract

The present invention discloses an early detection method for eggplant Verticillium wilt based on multispectral imaging technology, comprising the following steps: S1, using an artificial inoculation method simulating a natural environment to cultivate eggplant seedlings with Verticillium wilt disease characteristics, and using healthy eggplant seedlings of the same growth period as a control; S2, collecting multispectral images and chlorophyll fluorescence images of complete eggplant leaves of different growth days. The present invention adopts the above-mentioned early detection method for eggplant Verticillium wilt based on multispectral imaging technology, and uses multispectral imaging technology to classify eggplant leaves infected with Verticillium wilt before symptoms appear, specifically into three stages: healthy leaves, leaves about to be infected, and infected leaves; multispectral information is fused to establish an early detection model for eggplant leaf Verticillium wilt, and the stage is determined according to the multispectral information of the eggplant leaves, so as to make targeted treatments on eggplant plants, which can reduce the losses caused by Verticillium wilt to a greater extent.

Description

Technical Field

[0001] The invention relates to the technical field of early detection of Verticillium wilt, and in particular to an early detection method of eggplant Verticillium wilt based on multispectral imaging technology. Background Art

[0002] Early detection and treatment of Verticillium wilt is of great significance to ensure eggplant yield and quality. At present, eggplant Verticillium wilt is mainly controlled by chemical control, physical control and agricultural measures. However, these control methods are not effective against soil-borne fungal diseases.

[0003] Traditional manual diagnosis methods mainly rely on plant protection personnel to monitor the disease by observing eggplant leaves based on experience or combining laboratory testing methods. These traditional disease monitoring methods are time-consuming and labor-intensive, with low diagnostic efficiency, which limits the rapid detection and prevention of Verticillium wilt.

[0004] Spectroscopic technology can detect external characteristics, internal structure and other physiological indicators of plants. It has the advantages of real-time, non-contact, accurate and rich in information, and is widely used in information monitoring.

[0005] Gu et al. used hyperspectral imaging technology to make early diagnosis of tobacco tomato spotted wilt virus disease. The continuous projection algorithm combined with enhanced regression tree classification achieved the best results, with an average accuracy of 85.2%. They also found that the near-infrared spectral region is of great significance for early diagnosis of the disease, demonstrating the advantages of near-infrared spectroscopy technology in early diagnosis of the disease.

[0006] Kang et al. used drones to collect cotton canopy images and spectral data combined with convolutional neural networks to assess the severity of cotton Verticillium wilt, achieving both spatial and spectral disease severity assessments. However, these technologies are only used to distinguish whether plants are healthy or not, and do not predict plants that are about to become sick. Summary of the invention

[0007] The purpose of the present invention is to provide an early detection method for eggplant Verticillium wilt based on multispectral imaging technology, which can detect eggplant Verticillium wilt before symptoms appear, thereby minimizing the loss caused by Verticillium wilt.

[0008] To achieve the above object, the present invention provides an early detection method for eggplant Verticillium wilt based on multispectral imaging technology, comprising the following steps:

[0009] S1. Cultivating eggplant seedlings with Verticillium wilt disease characteristics by artificial inoculation method simulating natural environment, and using healthy eggplant seedlings at the same growth stage as control;

[0010] S2, collecting multispectral images and chlorophyll fluorescence images of complete eggplant leaves at different growth days;

[0011] S3, preprocessing the multispectral image obtained in step S2, and extracting characteristic bands;

[0012] S4, dividing the eggplant seedling samples into three categories: healthy leaves, leaves about to be infected, and infected leaves according to the multispectral image information pre-processed in step S3;

[0013] S5, using the multispectral image and / or chlorophyll fluorescence image of the characteristic band as the input of the CNN model, and using the healthy leaves, the leaves about to be infected and the infected leaves as the output, to train the CNN model and establish an early diagnosis model for eggplant leaf Verticillium wilt;

[0014] S6. Apply the early diagnosis model of eggplant leaf Verticillium wilt to eggplant plants to achieve early detection of eggplant leaf Verticillium wilt.

[0015] Preferably, the step S1 of cultivating eggplant seedlings with Verticillium wilt disease characteristics comprises the following specific steps:

[0016] S1-1. Select several eggplant seedlings with 2-4 true leaves, remove the seedlings from the plug tray, wash the soil around the roots with sterile water, and perform root cutting;

[0017] S1-2, soak the roots of eggplant seedlings to a concentration of 1×10 7 The inoculation was carried out in a bacterial solution with spores / mL, and the inoculation time was 10 minutes;

[0018] S1-3, planting the inoculated eggplant seedlings in a diseased environment at a constant temperature of 25°C for acclimatization for 1-2 days;

[0019] S1-4. Continue to cultivate the eggplant seedlings that have completed seedling acclimatization until the eggplant seedlings are observed to have typical characteristics of Verticillium wilt disease, and observe the leaves under a microscope to confirm the presence of Verticillium dahliae.

[0020] Preferably, the multispectral image processing in S3 includes:

[0021] S2-1, extracting spectral reflectance, chlorophyll fluorescence parameters and image texture feature data from multispectral images;

[0022] S2-2, eliminating the outliers of the data obtained in S2-1 according to the Mahalanobis distance;

[0023] S2-3, using the train_test_split() function in Sk-learn, divide all eggplant leaves into training set and test set according to 8:2;

[0024] S2-4, pre-process the data obtained in S2-1 using the normalization method, the normalization formula is:

[0025]

[0026] Where x is the original spectral data, μ is the mean of the original spectral data, x′ is the normalized data, and x max is the maximum value in the original spectral data, x min is the minimum value in the original spectral data.

[0027] Preferably, step S4 specifically comprises: using principal component analysis to analyze the spectral reflectance information extracted from the multispectral image, and dividing the leaves into three categories: healthy leaves, leaves about to be diseased, and diseased leaves.

[0028] Preferably, before training the CNN model in step S5, the original data of the CNN model is trained using the sparrow optimization algorithm to screen the optimal values ​​of the learning rate, batch processing quantity, and regularization coefficient.

[0029] Preferably, in step S5, the input of the CNN model is the spectral reflectance of the preprocessed characteristic band, and the CNN model is a CNN-SVM model.

[0030] Preferably, in step S5, the input of the CNN model is the spectral reflectance of the characteristic band, the chlorophyll fluorescence parameter and the image texture feature fusion, and the CNN model is a CNN-SVM model.

[0031] Preferably, in step S5, the input of the CNN model is a fusion of a multispectral image and a chlorophyll fluorescence image, and the CNN model is a 2D-CNN model.

[0032] Therefore, the present invention adopts the above-mentioned early detection method of eggplant Verticillium wilt based on multispectral imaging technology, and uses multispectral imaging technology to classify eggplant leaves infected with Verticillium wilt before symptoms appear, specifically dividing them into three stages: healthy leaves, leaves about to be infected, and leaves infected; multispectral information is fused to establish an early detection model for eggplant leaf Verticillium wilt, and the stage is determined according to the multispectral information of the eggplant leaves, so that targeted treatment can be made for the eggplant plants, which can reduce the losses caused by Verticillium wilt to a greater extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 The characteristic band selected by the continuous projection algorithm in the first embodiment;

[0034] Figure 2 is the average spectral reflectance of eggplant leaves from 0 day to 13 days in Example 1;

[0035] Figure 3 This is a schematic diagram of the structure of 1D-CNN and 2D-CNN models in Example 1;

[0036] Figure 4 This is a schematic diagram of the results of analyzing the spectral reflectance of healthy leaves and diseased leaves using the principal component analysis method in Example 1. DETAILED DESCRIPTION

[0037] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0038] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.

[0039] Embodiment 1

[0040] An early detection method for eggplant Verticillium wilt based on multispectral imaging technology specifically comprises the following steps:

[0041] S1. Preparation of Verticillium dahliae solution

[0042] The Verticillium wilt fungus blocks (isolated and purified from the main eggplant producing areas in Hebei Province) were cultured in potato glucose liquid culture medium.

[0043] The culture medium formula is: take 32g of potato glucose broth and 1000ml of distilled water. After mixing the solution evenly, place it in a sterilizer at 121℃ for 20 minutes. Select a small piece of Verticillium wilt bacteria and place it in the culture medium (this work is completed in a clean bench), save it with a sealing film, and place it in a 28° shaker for about 10-15 days. When preparing the bacterial solution, filter it with three layers of gauze and place it in a 50ml conical tube at 4000r / min for 15min. Pour out the supernatant after centrifugation, pour two-thirds of distilled water into the tube, and use a hemocytometer to adjust the concentration to 1×10 7 Spores / mL.

[0044] S2. Eggplant Plant Preparation

[0045] The high-generation inbred line "11-435" was selected as the experimental material. Eggplant seedlings were cultivated in a special breeding environment. When the seedlings grew 2-4 true leaves, the seedlings were taken out of the plug tray, the soil around the roots was washed with sterile water, and the roots were properly cut off. A total of 240 seedlings were obtained. The roots of half of the eggplant seedlings were soaked in a concentration of 1×10 7 The inoculation was carried out in a bacterial solution with spores / mL for 10 minutes. Then the seedlings were allowed to grow slowly for 1-2 days in a constant temperature of 25°C in a diseased environment. The roots of the other half of the eggplant seedlings were soaked in clean water for 10 minutes.

[0046] S3. Plant disease confirmation

[0047] Fifteen days after inoculation, it was found that all eggplant seedlings that were artificially inoculated had typical characteristics of Verticillium wilt disease, and leaves were taken and observed under a microscope to confirm the presence of Verticillium dahliae.

[0048] S4. Multispectral imaging

[0049] The multispectral imaging device VideometerLab 4 (Denmark) was used to collect the multispectral images of the eggplant leaf samples in a closed diffusion sphere, and finally a complete multispectral image of the eggplant leaf was obtained.

[0050] S5. Multispectral imaging data processing

[0051] S5-1. Information extraction of multispectral images includes: spectral reflectance, chlorophyll fluorescence parameters (CHIF) and image texture features.

[0052] Chlorophyll fluorescence parameters (CHIF) are often used to quantify the effects of environmental stress on plant photosynthesis. When plants are infected by Verticillium wilt, photosynthesis and chlorophyll content will change, changing the complexity of the CHIF signal in the time and frequency domains. Therefore, CHIF can be used to diagnose Verticillium wilt stress.

[0053] Texture features describe the surface properties of the scene corresponding to an image or an image region. It is a value calculated from the image that quantifies the characteristics of regional changes. Texture analysis is a process of extracting texture features through certain image processing techniques to obtain a quantitative or qualitative description of the texture.

[0054] The present invention adopts GLCM and LBP to extract texture features of multispectral images, derives contrast, correlation, energy and homogeneity from GLCM, extracts LBP1 to LBP59 parameters from LBP, removes parameters with a value of 0, and leaves a total of 16 parameters.

[0055] Generally speaking, the deeper the grooves of the contrast texture, the greater the contrast and the clearer the visual effect; conversely, if the contrast is small, the grooves are shallow and the effect is blurred.

[0056] The correlation reflects the local grayscale correlation in the image.

[0057] Energy is an expression of the uniformity of grayscale distribution and the fineness of texture.

[0058] Homogeneity measures the local uniformity of an image, with lower values ​​for non-uniform images and higher values ​​for uniform images.

[0059] LBP (Local Binary Pattern) is an operator used to describe the local texture features of an image. It has the characteristics of multi-resolution, grayscale invariance, rotation invariance, etc. It is mainly used for texture extraction in feature extraction.

[0060] S5-2. Data outlier detection

[0061] For the spectral reflectance extracted by S5-1, the Mahalanobis distance of each data is calculated, and the outliers exceeding the set threshold (7373279) are directly eliminated; the Mahalanobis distance d 2 The expression of (ΔtG) is:

[0062]

[0063] Where G is the spectral reflectance, μ is the mean vector, σ 2 is the covariance, and x represents the data to be tested.

[0064] In this embodiment, there are a total of 1431 samples, and 1392 samples remain after data removal.

[0065] S5-3. Data Division

[0066] Use the train_test_split() function in Sk-learn to divide the eggplant leaves into training set and test set in the ratio of 8:2.

[0067] S5-4. Data preprocessing

[0068] Due to the large differences in spectral reflectance and image texture features, normalization is used to preprocess the data. This method converts the original data into dimensionless values ​​at the same level of standard to facilitate accurate comprehensive analysis and evaluation of all data. The normalization formula is:

[0069]

[0070] Where x is the original data, μ is the mean of the original data, x′ is the normalized data, and x max is the maximum value in the original data, x min is the minimum value in the original data.

[0071] S5-5. Feature band extraction

[0072] Extracting characteristic bands can reduce the amount of model calculations, improve the model's calculation speed, and reduce the probability of model overfitting. The present invention uses a continuous projection algorithm (SPA) to extract characteristic bands. The continuous projection algorithm projects band information onto other wavelength information through mathematical transformation, and the characteristic band is the band information with the largest projection.

[0073] Assume the number of samples M and the number of wavelengths K as the spectral absorption matrix XM*K, with X K(0) and N (N≤M-1) are the initial iteration vector and the number of wavelengths set. The steps are as follows:

[0074] (1) n = 1, select a vector X in the spectrum matrix K(0) ;

[0075] (2) The set S is defined as Calculate x separately j Projection vector of S vector:

[0076]

[0077] (3) Maximum number of records:

[0078] k(n)=arg(max||Px j ||, j∈S)

[0079] (4) Take the maximum projection as the vector x for the next iteration j =Px j , j∈S,

[0080] (5) n = n + 1, if n < N, continue the loop, otherwise exit the loop.

[0081] After the continuous projection algorithm was used to reduce the dimension of the data, 10 optimal bands were selected: 430nm, 490nm, 515nm, 540nm, 570nm, 645nm, 660nm, 780nm, 850nm, and 970nm. Among them, 970nm is usually related to water content, while 850nm usually represents protein and lipid content. In addition, 430, 490, 515, 540, and 570nm are related to chlorophyll content, and 645nm and 660nm are related to leaf nitrogen content. The characteristic bands selected by the continuous projection algorithm are as follows: Figure 1 shown.

[0082] S5-6. Spectral data analysis

[0083] Figure 2 The average spectral reflectance of eggplant leaves from 0 day to 13 days is shown.

[0084] Healthy leaves and diseased leaves have similar spectral reflectance trends, but the spectral reflectances of 0day, 3days, and 5days to 11days are different, with large differences in the 500-570nm band and the near-infrared band. 500-570nm is the green light band. When eggplant leaves are infected with Verticillium wilt, the leaves will gradually turn yellow, resulting in changes in spectral reflectance. The near-infrared band is related to the moisture content of the leaves. Therefore, compared with healthy leaves, the physiological changes caused by the diseased leaves lead to different spectral reflectances.

[0085] The principal component analysis method was used to analyze the spectral reflectance of healthy leaves and diseased leaves, and the 19 bands were converted into two comprehensive variables PC 1 、PC 2 , Figure 4 Showing PC 1 、PC 2 Scatter plot of scores. PC 1 、PC 2 Contains most of the spectral information, accounting for 95.6% of the total variance, of which PC 1 Accounting for 83.4% of the explained variance, PC 2 The score chart shows the classification of healthy leaves and infected leaves. According to observation, as the disease progresses, it can be divided into three categories: 0day, 3days, and 5days to 11days, namely healthy leaves, leaves about to be infected, and leaves infected.

[0086] S5-6. Feature fusion and recognition model

[0087] The sparrow optimization algorithm was used to train the original data of the CNN model, and the number of iterations was set to 5 and the population size was set to 20. Finally, three optimization parameters were selected, namely the learning rate, the number of batch processing, and the regularization coefficient. Before training the early diagnosis model of eggplant Verticillium wilt, the lower bounds of the three parameters were set to 1e-3, 32, and 1e-5, and the upper bounds of the three parameters were set to 5e-2, 128, and 1e-2, respectively. Different optimization parameters were obtained according to different data types and input into the CNN model.

[0088] CNN-SVM refers to the use of a convolutional neural network trained model to extract high-dimensional features, ultimately obtaining three-dimensional data, substituting the three-dimensional characteristic data into the SVM, and training the classification model again to improve the classification accuracy.

[0089] The basic structure of CNN consists of input layer, convolutional layer, pooling layer, fully connected layer and output layer.

[0090] Based on the spectral information and image information of eggplant leaves, 1D-CNN and 2D-CNN models are proposed. 1D-CNN has two convolutional layers, two maximum pooling layers, and one fully connected layer, and finally softmax is used to complete disease diagnosis.

[0091] To reduce overfitting, a dropout layer is added before each fully connected layer. To overcome the gradient vanishing problem and speed up training, ReLU is used as the activation function unit.

[0092] 2D-CNN is composed of an input layer, three convolutional layers, three activation layers corresponding to the convolutional layers, three pooling layers corresponding to the convolutional layers, and a fully connected layer. The input layer is a 100×100×19 multispectral image. Figure 3 The structural diagram of 1D-CNN and 2D-CNN

[0093] The reflectance of the full band, the reflectance of the characteristic band and the reflectance of the characteristic band after preprocessing were used as the input of the model, and the healthy leaves, the leaves about to be diseased and the diseased leaves were used as the model output. 1D-CNN, CNN-SVM, SSA-CNN and SSA-CNN-SVM machine learning algorithms were used for modeling.

[0094] The modeling results are shown in Table 1. According to the results, the spectral data processed by preprocessing and feature band extraction has the best classification effect using CNN-SVM, and its test set accuracy has reached 99.12%. At the same time, the calculation speed of the model is also improved due to the small number of input bands. The results show that spectral data is feasible for early diagnosis of eggplant Verticillium wilt.

[0095] Table 1 Early diagnosis classification results with spectral features

[0096]

[0097] Embodiment 2

[0098] The multispectral imaging data collected in Example 1 is continued to be used, except that spectral data combined with chlorophyll fluorescence parameters and image texture features are used as the model input for early diagnosis of eggplant leaf Verticillium wilt.

[0099] Specifically, the reflectance of the full band combined with chlorophyll fluorescence parameters and image texture features, the reflectance of the characteristic band combined with chlorophyll fluorescence parameters and image texture features, and the reflectance of the characteristic band after preprocessing combined with chlorophyll fluorescence parameters and image texture features are used as the model input. 12 machine learning models were constructed based on the fusion features, and Table 2 shows the model performance based on the fusion features.

[0100] Table 2. Early diagnosis classification results with spectral features, image texture features, and chlorophyll fluorescence parameters

[0101]

[0102] Compared with the spectral feature model in Example 1, the model integrating texture features generally has better effects, among which the SPA combined with the CNN-SVM model has the best effect, with the accuracy of the training set reaching 100% and the test set result reaching 99.12%.

[0103] Therefore, spectral data combined with chlorophyll fluorescence parameters and image texture features can more fully explain the changes in the internal composition and external properties of leaves during the infection process, and is more advantageous for the early diagnosis of eggplant leaf wilt.

[0104] Embodiment 3

[0105] The multispectral imaging data collected in Example 1 is continued to be used, except that multispectral images and multispectral images combined with chlorophyll fluorescence images are used as inputs of the 2D-CNN model to perform early diagnosis of eggplant leaf wilt.

[0106] The multispectral image contains images under 19 different wavelengths of LED flashes, including spectral reflectance, image texture and other features. Chlorophyll fluorescence is measured by 700LP emission filter through illumination-induced chlorophyll excitation at 630nm, 645nm and 660nm. After measurement, chlorophyll fluorescence images at 365nm, 405nm, 430nm, 450nm, 470nm and 490nm are obtained.

[0107] Table 3 shows the model comparison results, where 25-images is the fusion model of multispectral images and chlorophyll fluorescence images, 19-images is the multispectral image model, and 11-images is the multispectral image and chlorophyll fluorescence image characteristic band fusion model (470nm, 515nm, 570nm, 645nm, 660nm, 780nm, 850nm, 880nm, 940nm, 970nm, c6 (the sixth photo of chlorophyll fluorescence image is referred to as C6)).

[0108] Table 3. Early diagnosis classification results of multispectral images and multispectral images combined with chlorophyll fluorescence images

[0109]

[0110] The results show that the multispectral and chlorophyll fluorescence images after feature band extraction using the continuous projection algorithm performed best, with a validation set accuracy of 99.37%, which shows that images based on fusion features perform better than single feature models and fusion feature data.

[0111] Embodiment 4

[0112] The experiment was conducted using the round eggplant inbred line 14-345 as a model. The eggplant seedlings were artificially inoculated with fungi using the same method as in Example 1. Multispectral images and chlorophyll fluorescence image information were collected every day after inoculation. The images were input into the early diagnosis model of Verticillium wilt in Example 3 after transfer learning. The results are shown in Table 4.

[0113] Table 4. Early diagnosis classification results

[0114]

[0115] Therefore, the present invention adopts the above-mentioned early detection method of eggplant Verticillium wilt based on multispectral imaging technology, and uses multispectral imaging technology to classify eggplant leaves infected with Verticillium wilt before symptoms appear, specifically dividing them into three stages: healthy leaves, leaves about to be infected, and leaves infected; multispectral information is fused to establish an early detection model for eggplant leaf Verticillium wilt, and the stage is determined according to the multispectral information of the eggplant leaves, so that targeted treatment can be made for the eggplant plants, which can reduce the losses caused by Verticillium wilt to a greater extent.

[0116] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. An early detection method for eggplant Verticillium wilt based on multispectral imaging technology, characterized in that: The steps include: S1. Cultivating eggplant seedlings with Verticillium wilt disease characteristics by artificial inoculation method simulating natural environment, and using healthy eggplant seedlings at the same growth stage as control; S2, collecting multispectral images and chlorophyll fluorescence images of complete eggplant leaves at different growth days; S3, preprocessing the multispectral image obtained in step S2, and extracting characteristic bands; S4, dividing the eggplant seedling samples into three categories: healthy leaves, leaves about to be infected, and infected leaves according to the multispectral image information pre-processed in step S3; S5, taking the multispectral image and chlorophyll fluorescence image of the characteristic band as the input of the CNN model, taking the healthy leaves, the leaves about to be infected and the infected leaves as the output, training the CNN model, and establishing an early diagnosis model for eggplant leaf Verticillium wilt; S6. Applying the early diagnosis model of eggplant leaf Verticillium wilt to eggplant plants to achieve early detection of eggplant leaf Verticillium wilt; Multispectral image preprocessing in S3 includes: S2-1, extracting spectral reflectance, chlorophyll fluorescence parameters and image texture feature data from multispectral images and chlorophyll fluorescence images; S2-2, eliminating the outliers of the data obtained in S2-1 according to the Mahalanobis distance; S2-3, using the train_test_split() function in Sk-learn to divide all eggplant leaves into training set and test set according to 8:2; S2-4, pre-process the data obtained in S2-1 using the normalization method, the normalization formula is: ； in is the original data, To represent the original data mean, is the normalized data, is the maximum value in the original data, is the minimum value in the original data; Step S4 specifically includes: using principal component analysis to analyze the spectral reflectance information extracted from the multispectral image, and classifying the leaves into three categories: healthy leaves, leaves about to be infected, and leaves infected; Before the CNN model training in step S5, the original data of the CNN model is trained using the sparrow optimization algorithm to select the optimal values ​​of the learning rate, the batch processing quantity, and the regularization coefficient; the input of the CNN model is a fusion of a multispectral image and a chlorophyll fluorescence image, and the CNN model is a 2D-CNN model.

2. The method for early detection of eggplant Verticillium wilt based on multispectral imaging technology according to claim 1, characterized in that: The step S1 of cultivating eggplant seedlings with Verticillium wilt disease characteristics specifically comprises the following steps: S1-1. Select several eggplant seedlings with 2-4 true leaves, remove the seedlings from the plug tray, wash the soil around the roots with sterile water, and perform root cutting; S1-2, soak the roots of eggplant seedlings to a concentration of 1×10 7 The inoculation was carried out in a bacterial solution with spores / mL, and the inoculation time was 10 minutes; S1-3, planting the inoculated eggplant seedlings in a diseased environment at a constant temperature of 25°C for acclimatization for 1-2 days; S1-4. Continue to cultivate the eggplant seedlings that have completed seedling acclimatization until the eggplant seedlings are observed to have typical characteristics of Verticillium wilt disease, and observe the leaves under a microscope to confirm the presence of Verticillium dahliae.

Images