Cotton verticillium wilt early detection method based on hyperspectral imaging recurrence plot

Abstract

The invention provides a cotton verticillium wilt early detection method based on a hyperspectral imaging recurrence plot, which comprises the following steps: inoculating cotton verticillium wilt germs to obtain healthy plants, non-symptomatic plants and symptomatic plants; acquiring and processing hyperspectral images in different verticillium wilt states; converting the one-dimensional hyperspectral curve into a two-dimensional recurrence plot (RP); performing graying processing on the two-dimensional recursive image; extracting a gray gradient co-occurrence matrix (GLGCM) of different images; carrying out Pearson correlation analysis, and evaluating the correlation between the texture features and the three types of labels; eliminating redundant information of the features by adopting principal component analysis (PCA), and obtaining a load matrix of each principal component; based on the four machine learning models, a three-classification cotton plant verticillium wilt early detection model of'healthy, non-symptom-symptom 'is established. According to the method, the feasibility of converting the spectrum curve into the recursion plot and extracting the image texture features for accurate identification of the cotton verticillium wilt in the asymptomatic period is proved, and reliable reference is provided for accurate identification of cotton verticillium wilt infection.

Description

Technical Field

[0001] The invention relates to the technical field of crop disease diagnosis, and in particular to an early detection method for cotton Verticillium wilt based on a hyperspectral imaging recursive graph. Background Art

[0002] Due to the continuous cropping of cotton, unreasonable management methods and natural factors, cotton diseases have become more serious in recent years. Among them, the annual incidence area of ​​cotton Verticillium wilt accounts for 36.6%, and the annual cotton loss rate is over 60%, causing serious economic losses and yield losses. Cotton Verticillium wilt is a soil-borne fungal disease caused by Verticillium dahliae. The microsclerotia of the pathogen can exist in the soil for a long time and can resist relatively harsh environments. So far, there is no large-scale and accurate method for identifying Verticillium dahliae infection in the asymptomatic period of cotton. In order to achieve the degree of infection of cotton Verticillium wilt, a method for early detection of cotton Verticillium wilt using hyperspectral imaging and recursive graph combined with machine learning technology is urgently needed.

[0003] Cotton is very susceptible to diseases during its growth process, which seriously affects its yield and quality, causing an average annual yield loss of 15%-20%. Among them, Verticillium wilt is the most serious and widely distributed cotton disease, known as "cotton cancer". When cotton suffers from more serious Verticillium wilt, it is easy to cause a reduction in production or even a total crop failure, causing significant economic and social losses. Therefore, rapid and accurate monitoring of cotton Verticillium wilt is of great significance for the comprehensive prevention and control of Verticillium wilt.

[0004] Traditional detection methods still mainly rely on visual inspection to confirm the type and extent of the disease, which has low accuracy and lags behind. Indoor fungal culture detection of Verticillium dahliae can achieve accurate early diagnosis, but this method requires destructive sampling and is time-consuming, labor-intensive and expensive. At present, the application of spectral signal imaging methods based on recursive graphs in the early detection of cotton Verticillium wilt is relatively rare.

[0005] In view of this, the present invention provides an early detection method for cotton Verticillium wilt based on hyperspectral imaging recursive graph to solve the above-mentioned problems existing in the prior art. Summary of the invention

[0006] To achieve the above object, the present invention proposes an early detection method for cotton Verticillium wilt based on hyperspectral imaging recursive graph, which converts spectral curves into two-dimensional recursive graphs for early detection of cotton Verticillium wilt and accurate diagnosis, and specifically includes the following steps:

[0007] Two cotton varieties with different resistance to Verticillium wilt were selected and cotton was planted in hydroponic planting mode.

[0008] The inoculation mode of bacterial vacuole roots was adopted to complete the inoculation of cotton Verticillium wilt pathogen and obtain healthy plants, non-symptomatic plants and symptomatic plants.

[0009] The hyperspectral imaging system is used to collect, process and analyze the spectra of cotton plants in different states.

[0010] The hyperspectral curves of different Verticillium wilt states were converted into two-dimensional recursion plot (RP) images.

[0011] The recursive graph image is grayed out and 15 gray-level gradient co-occurrence matrices (GLGCM) texture features are further extracted.

[0012] Pearson correlation analysis of the recursive graph texture features was carried out to evaluate the correlation between the texture features and the three types of labels.

[0013] The principal component dimensionality reduction process is performed on the screened texture feature parameters to eliminate the redundant information of the features.

[0014] A machine learning model for early detection of Verticillium wilt in cotton plants was constructed based on principal component analysis with dimensionality reduction, and the classification performance of each model was compared and analyzed to select the detection model with the best performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the following detailed description of the preferred embodiment.The drawings are only for the purpose of illustrating the preferred embodiments and are not to be construed as limiting the invention.

[0016] In the attached picture: Figure 1 Schematic diagram of cotton planting and Verticillium wilt pathogen inoculation in an embodiment of the present invention; Figure 2 Schematic diagrams of cotton plants in different Verticillium wilt infection states according to an embodiment of the present invention, wherein Figure A is an image of a healthy cotton plant, Figure B is an image of a cotton plant without Verticillium wilt symptoms, and Figure C is an image of a cotton plant with Verticillium wilt symptoms; Figure 3 Schematic diagram of a hyperspectral imaging system for cotton Verticillium wilt in an embodiment of the present invention; Figure 4 Schematic diagram of typical spectral curves of cotton plants in three states in an embodiment of the present invention; Figure 5 Schematic diagram of two-dimensional recursive graph images of spectral curves of cotton plants in three states according to an embodiment of the present invention, wherein Figure A is a recursive graph image of a healthy plant, Figure B is a recursive graph image of a plant without symptoms of Verticillium wilt, and Figure C is a recursive graph image of a plant with symptoms of Verticillium wilt; Figure 6 Schematic diagram of the correlation analysis results between the recursive graph texture features and three category labels in an embodiment of the present invention; Figure 7is a schematic diagram of the correlation coefficients between the texture features after screening in an embodiment of the present invention; Figure 8 Schematic diagram of the characteristic value and variance contribution rate of each principal component in an embodiment of the present invention; Fig. 9 Schematic diagram of confusion matrix of different machine learning classification models in the embodiments of the present invention, wherein Figure A is a K-nearest neighbor (KNN) model, Figure B is a support vector machine (SVM) model, Figure C is an extreme learning machine (ELM) model, and Figure D is an extreme gradient boosting machine (XGBoost) model; Fig.10 Schematic diagram of comparative analysis of the performance of different machine learning models for Verticillium wilt detection in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present disclosure and to be able to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, in the absence of conflict, the embodiments of the present invention and the features described in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0018] This embodiment proposes a method for early detection of cotton Verticillium wilt based on hyperspectral imaging recursive graph. Figure 1 As shown, the following steps are included.

[0019] Step 1: Select two cotton varieties with different resistance to Verticillium wilt and plant them in a hydroponic planting mode.

[0020] Specifically, two cotton varieties with different resistance to Verticillium wilt were selected, including Xinlu Zao 76 and Xinlu Zao 1. This increased the diversity of data samples and covered cotton with different levels of resistance to Verticillium wilt. The hydroponic planting mode was selected for cotton planting to reduce the impact of external factors and more accurately control the growth of cotton. The hydroponic nutrient solution used was the Hogland standard concentration nutrient solution formula, which is a universal standard nutrient solution for soilless cultivation.

[0021] Step 2: Use the bacterial bubble root inoculation mode to complete the inoculation of cotton Verticillium wilt pathogens and obtain healthy plants, non-symptomatic plants and symptomatic plants.

[0022] Specifically, in order to improve the success rate of Verticillium wilt inoculation, the inoculation mode of bacterial vacuole root was adopted. The activated cultured mycelium of Verticillium dahliae on potato dextrose agar medium (PDA) was taken and cultured in complete medium (CM) liquid medium; then the culture solution was filtered with 4 layers of gauze, the mycelium was removed by vacuum filtration, the filtrate was retained, and the spare filtrate was centrifuged at 10000 r / min for 30 min to separate the required precipitate. After removing the supernatant, the precipitate was resuspended in 2 mL of sterile water. Finally, the concentration of cotton Verticillium wilt spores was prepared with sterile water to obtain 1×10 7 mol / L spore suspension of Verticillium wilt. When the cotton has 4 leaves and 1 core, choose the right time to inoculate. Soak the roots of the cotton to be infected in the prepared cotton Verticillium wilt solution, let it stand for 30 minutes, and then immediately put the inoculated cotton back into the hydroponic box to complete the inoculation of cotton Verticillium wilt.

[0023] Furthermore, for cotton plants under different infection conditions, the infection process of Verticillium wilt is divided into two stages, namely the asymptomatic stage of Verticillium wilt and the symptomatic stage of Verticillium wilt. Healthy plants have no diseased leaves, grow normally, and have no lesions in the xylem of the stems. The appearance of the leaves of cotton plants infected with asymptomatic disease is not much different from that of healthy plants, but the edges and interior of the leaves are slightly yellow, and the xylem of the stems is diseased. At this time, in the vascular tissue of the plant, the hyphae of Verticillium dahliae produce spores and spread rapidly in the vascular bundles. Cotton plants infected with symptoms of Verticillium wilt are significantly different from healthy plants. The leaves are diseased and turn yellow and wilt, and the vascular bundles show large areas of browning. At this time, Verticillium dahliae degrades the root cell walls of crops by secreting toxins, and parasitizes in the xylem of plants in large numbers. As water evaporates, it spreads toward the stems and leaves of the plants, and eventually infects the entire crop.

[0024] Step 3: Hyperspectral image acquisition and processing.

[0025] Specifically, a continuous shooting strategy was adopted, combined with the experience of agricultural experts and the national standard GB / T 22101, to identify each cotton plant until the cotton showed symptoms of Verticillium wilt. The sample on the day of onset was the captured early sample of Verticillium wilt. The hyperspectral imaging system mainly consists of two modules: hardware and software. The hardware part mainly includes SOC710VP hyperspectral imager, light source, height adjuster, diffuse reflection reference plate, dark box, and transmission line. The software part uses HyperScanner hyperspectral image data acquisition tool and SRAnalysis data analysis tool compatible with SOC equipment. The complete cotton plant was defined as the region of interest (ROI), and the threshold was set to 0.02 to distinguish the foreground ROI from the background. A binary mask was created accordingly, in which 0 represents the background and 1 represents the foreground cotton plant. Finally, the mask was applied to the complete hyperspectral image to obtain a hyperspectral image with the background removed. Then, the average value of the pixels in each channel was calculated to preliminarily obtain the cotton spectral curve with the background removed. Due to the presence of dark current in the CCD camera and the uneven illumination intensity in different bands, some bands with weaker light intensity contain larger noise. In order to remove the noise caused by the environment and dark current, the original hyperspectral image is calibrated in black and white.

[0026] Preferably, the difference between the spectral curves of healthy samples and non-symptomatic samples is small, the spectral reflectance value of healthy samples is high, and the spectral reflectance value of infected samples gradually decreases as the number of days of infection increases. The spectral profile of cotton plants infected with Verticillium wilt without symptoms is slightly different from that of healthy cotton plants. The microscopic internal cell atrophy of cotton plants with symptomatic Verticillium wilt causes macroscopic cotton leaf withering, the plant moisture content decreases, and the near-infrared spectral reflectance value decreases significantly.

[0027] Step 4: The hyperspectral curves of different Verticillium wilt states were converted into recursion plots (RP).

[0028] Specifically, the hyperspectral time series curves of cotton plants in different Verticillium wilt states were taken as the research objects, and the one-dimensional spectral signal time series was reconstructed into a high-dimensional phase space and converted into a two-dimensional image.

[0029] Takens embedding theorem is used to reconstruct the phase space of the original time series. The specific formula is:

[0030] In the formula, is the delay time, d is the embedding dimension, The corresponding delay time and embedding dimension are obtained by mutual information analysis and false nearest neighbor algorithm. .

[0031] Select the distance between any two vectors in Euclidean space, the specific formula is:

[0032] Depending on the selected distance threshold ε, the recurrence graph is represented as:

[0033] In the formula, is the Heavside function, and ε is usually set to 10% of the standard deviation of the original time series.

[0034] Furthermore, there are obvious texture differences in the two-dimensional recursive graphs of the spectral curves of different types of samples. The color mapping intensity inside the recursive image of healthy plants is high, and the color mapping intensity inside the recursive image of infected plants without symptoms gradually weakens, which is related to the decrease in the reflectivity of the spectral curve of non-symptomatic plants relative to the reflectivity of the spectral curve of healthy plants. The color mapping intensity inside the recursive image of infected plants with symptoms drops significantly, and the reflectivity of the corresponding spectral curve drops below 0.4.

[0035] Step 5: The RP image is grayscaled and 15 gray-level gradient co-occurrence matrix (GLGCM) features of the image are extracted.

[0036] Specifically, the hyperspectral curve RP images of cotton plants in three different states were grayed, and then the gray-level gradient co-occurrence matrix (GLGCM) was used to capture the second-order statistics of grayscale and gradient information in the grayscale recursive map image, and a total of 15 texture features were extracted.

[0037] Furthermore, different GLGCM feature parameters are used to characterize the texture distribution of the recursive image from different angles. The small gradient advantage, large gradient advantage, grayscale distribution unevenness, and gradient distribution unevenness describe the distribution of grayscale and gradient in the image texture respectively; the angular second-order moment reflects the uniformity and coarseness of the texture distribution; the grayscale mean and gradient mean characterize the concentration of grayscale and gradient in the image texture; the grayscale mean square error and gradient mean square error reflect the changing period of the image texture; the correlation describes the similarity of local grayscale and gradient in the image in rows or columns; the grayscale entropy, gradient entropy, and mixed entropy characterize the complexity or non-uniformity of the image texture; the moment of inertia characterizes the grooves and clarity of the image texture; and the inverse difference moment reflects the homogeneity and regularity of the image texture.

[0038] Step 6: Conduct Pearson correlation analysis to evaluate the correlation between texture features and the three types of labels.

[0039] Specifically, three types of labels were set for healthy cotton plants, non-symptomatic plants, and symptomatic plants. The correlation between 15 texture features and the three types of labels was evaluated through Pearson correlation analysis to determine the texture features that were moderately or strongly correlated with the occurrence of cotton Verticillium wilt.

[0040] Furthermore, the original color recursive images were converted into grayscale images, and 15 texture features were extracted using the GLGCM method. The absolute values ​​of the correlation coefficients of 11 texture features were higher than 0.6, showing strong correlations with the class labels of healthy, non-symptomatic, and symptomatic Verticillium wilt cotton plants. Among these features, five were related to entropy and asymmetry. For the remaining four texture features, two texture gradient features had moderate correlations with the three class labels (0.4<|r|<0.6), while the remaining two features, grayscale mean and grayscale asymmetry, were weakly correlated with the class labels (0.2<|r|<0.4). The four texture features were removed to reduce the complexity and computational cost of the input features.

[0041] Step 7: Use principal component analysis (PCA) to eliminate redundant information of features.

[0042] Specifically, the 11 texture feature parameters were processed by principal component dimensionality reduction, and the eigenvalues ​​of the first five principal components were greater than 1. The cumulative variance contribution rate of these five principal components was as high as 99.02%, which basically explained most of the information of the 11 texture feature parameters. Therefore, the first five principal components were determined to characterize the texture feature parameter information of the recursive graph of the spectrum of Verticillium wilt in different states of cotton.

[0043] Step 8: Construct a machine learning model for early detection of Verticillium wilt in cotton plants based on the principal component of dimensionality reduction, and compare and analyze the classification performance of each model to select the detection model with the best performance.

[0044] Furthermore, the K-nearest neighbor method (KNN), support vector machine (SVM), extreme learning machine (ELM) and extreme gradient boosting machine (XGBoost) were used to establish the "healthy-no symptoms-symptomatic" three-classification cotton plant Verticillium wilt detection model. The confusion matrix of the test set of the four models showed that in the KNN model, 7 healthy samples were misclassified as no symptoms, and 8 symptomatic samples were misclassified as no symptoms. In the SVM model, 6 healthy samples were misclassified as no symptoms, and 5 symptomatic samples were misclassified as no symptoms. In the ELM model, 6 healthy samples were misclassified as no symptoms, and 6 symptomatic samples were misclassified as no symptoms. In the XGBoost model, 3 healthy samples were misclassified as no symptoms, and 3 symptomatic samples were misclassified as no symptoms. By comparison, it can be seen that the XGBoost model has the least misclassification, followed by the SVM and ELM models, and the KNN model has the most misclassification.

[0045] Furthermore, the accuracy, precision, recall, and F1 scores were used to further compare and analyze the performance of different machine learning classification models. Among the four machine learning classification models, the KNN classification model had the lowest overall classification performance. The classification performance of the ELM and SVM classification models were improved compared to the KNN model, and the corresponding classification indicators increased. The accuracy, precision, recall, and F1 score of the XGBoost classification model were 96.3%, 95.6%, 96%, and 95.8%, respectively, showing the highest classification ability.

[0046] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can still modify the technical solutions described in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or perform equivalent replacements on some of the technical features thereof; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. They should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A method for early detection of cotton Verticillium wilt based on hyperspectral imaging recursive graph, characterized in that: The following steps are involved: Two cotton varieties with different resistance to Verticillium wilt were selected and cotton was planted in hydroponic planting mode; The inoculation mode of bacterial vacuole root was adopted to complete the inoculation of cotton Verticillium wilt pathogens, and healthy plants, non-symptomatic plants and symptomatic plants were obtained; The spectrum of cotton plants in different states is collected, processed and analyzed through a hyperspectral imaging system; The hyperspectral curves of different Verticillium wilt states were converted into two-dimensional recursive plot (RP) images; The recursive graph image is grayed out and 15 gray-level gradient co-occurrence matrix (GLGCM) texture features are further extracted. Conduct Pearson correlation analysis of the recursive graph texture features to evaluate the correlation between texture features and the three types of labels; The principal component dimension reduction process is performed on the filtered texture feature parameters to eliminate the redundant information of the features; A machine learning model for early detection of Verticillium wilt in cotton plants was constructed based on principal component analysis with dimensionality reduction, and the classification performance of each model was compared and analyzed to select the detection model with the best performance.

2. The method for early detection of cotton Verticillium wilt based on hyperspectral imaging recursive graph according to claim 1, characterized in that: The method comprises selecting two cotton varieties with different resistance to Verticillium wilt and planting cotton in a hydroponic planting mode. Two cotton varieties with different resistance to Verticillium wilt are selected, including Xinlu Zao 76 and Xinlu Zao 1. In order to increase the diversity of data samples and cover cotton with different levels of resistance to Verticillium wilt, a hydroponic planting mode is selected for cotton planting to reduce the influence of external factors and more accurately control the growth of cotton. The hydroponic nutrient solution adopts the Hogland standard concentration nutrient solution formula, which is a universal standard nutrient solution for soilless cultivation. The inoculation mode of bacterial vacuole root was adopted to complete the inoculation of cotton Verticillium wilt pathogen, and healthy plants, non-symptomatic plants and symptomatic plants were obtained. In order to improve the success rate of Verticillium wilt inoculation, the inoculation mode of bacterial vacuole root was adopted, and the activated cultured mycelium of Verticillium dahliae on potato dextrose agar medium (PDA) was taken and cultured in complete medium (CM) liquid medium; then the culture solution was filtered with 4 layers of gauze, the mycelium was removed by vacuum filtration, the filtrate was retained, and the spare filtrate was centrifuged at 10000 r / min for 30 min to separate the required precipitate. After removing the supernatant, the precipitate was resuspended with 2 mL of sterile water. Finally, the concentration of cotton Verticillium wilt pathogen spores was prepared with sterile water to obtain a 1×107 mol / L Verticillium wilt pathogen spore suspension. When the cotton has 4 leaves and 1 core, choose the right time to inoculate. Soak the roots of the cotton to be infected in the prepared cotton Verticillium wilt solution, let it stand for 30 minutes, and then immediately put the inoculated cotton back into the hydroponic box. This completes the inoculation of cotton Verticillium wilt. For cotton plants under different infection conditions, the infection process of Verticillium wilt is divided into two stages, namely the asymptomatic stage of Verticillium wilt and the symptomatic stage of Verticillium wilt. Healthy plants have no diseased leaves, grow normally, and have no lesions in the xylem of the stems. The leaves of asymptomatic cotton plants are slightly different from those of healthy plants in appearance, but the edges and inside of the leaves are slightly yellowed, and the xylem of the stems is lesioned. At this time, in the vascular tissue of the plant, the hyphae of Verticillium dahliae produce spores and spread rapidly in the vascular bundles. Cotton plants infected with Verticillium wilt and showing symptoms are significantly different from healthy plants. The leaves are diseased and turn yellow and wilt, and large areas of browning appear in the vascular bundles. At this time, Verticillium dahliae degrades the root cell walls of the crop by secreting toxins, and parasitizes in the xylem of the plant in large numbers. As water evaporates, it spreads toward the stems and leaves of the plant, eventually infecting the entire crop.

3. The method for early detection of cotton Verticillium wilt based on hyperspectral imaging recursive graph according to claim 1, characterized in that: The hyperspectral image acquisition and processing adopts a continuous shooting strategy, combined with the experience of agricultural experts and the national standard GB / T 22101 to identify each cotton plant until the cotton shows symptoms of Verticillium wilt. The sample on the day of the onset of the disease is the captured early cotton Verticillium wilt sample. The hyperspectral imaging system mainly includes two modules: hardware and software. The hardware part mainly includes SOC710VP hyperspectral imager, light source, height adjuster, diffuse reflectance reference plate, dark box, and transmission line. The software part uses HyperScanner hyperspectral image data acquisition tool and SRAnalysis data analysis tool compatible with SOC devices. The complete cotton plant is defined as the region of interest (ROI), and the threshold is set to 0.02 to distinguish the foreground ROI from the background. A binary mask is created accordingly, where 0 represents the background and 1 represents the foreground cotton plant. Finally, the mask is applied to the complete hyperspectral image to obtain a hyperspectral image with the background removed. Then the average value of the pixels is calculated for each channel to preliminarily obtain the cotton spectral curve with the background removed. Due to the dark current in the CCD camera and the uneven illumination intensity of different bands, some bands with weaker light intensity contain large noise. In order to remove the noise caused by the environment and dark current, the original hyperspectral image is calibrated in black and white; The difference between the spectral curves of healthy samples and non-symptomatic samples is small, and the spectral reflectance value of healthy samples is higher. As the number of days of infection increases, the spectral reflectance value of infected samples gradually decreases. The spectral profile of cotton plants infected with Verticillium wilt without symptoms is slightly different from that of healthy cotton plants. The microscopic internal cell atrophy of cotton plants with symptomatic Verticillium wilt causes macroscopic cotton leaf withering, the water content of the plants decreases, and the near-infrared spectral reflectance value decreases significantly.

4. The method for early detection of cotton Verticillium wilt based on hyperspectral imaging recursive graph according to claim 1, characterized in that: The hyperspectral curves of different Verticillium wilt states are converted into recursive graphs (RP), and the hyperspectral time series curves of cotton plants in different Verticillium wilt states are used as research objects, and the one-dimensional spectral signal time series is reconstructed into a high-dimensional phase space and converted into a two-dimensional image; Takens embedding theorem is used to reconstruct the phase space of the original time series. The specific formula is: In the formula, is the delay time, d is the embedding dimension, The corresponding delay time and embedding dimension are obtained by mutual information analysis and false nearest neighbor algorithm. Select the distance between any two vectors in Euclidean space, the specific formula is: Depending on the selected distance threshold ε, the recurrence graph is represented as: In the formula, is the Heavside function, ε is usually set to 10% of the standard deviation of the original time series There are obvious texture differences in the two-dimensional recursive images of the spectral curves of different types of samples. The color mapping intensity inside the recursive image of healthy plants is high, and the color mapping intensity inside the recursive image of infected plants without symptoms gradually weakens. This is related to the decrease in the reflectivity of the spectral curve of the non-symptomatic plant relative to the reflectivity of the spectral curve of the healthy plant. The color mapping intensity inside the recursive image of the infected symptomatic plant drops significantly, and the reflectivity of the corresponding spectral curve drops below 0.

4.

5. The method for early detection of cotton Verticillium wilt based on hyperspectral imaging recursive graph according to claim 1, characterized in that: The RP image is grayed out to extract 15 gray gradient co-occurrence matrix (GLGCM) features of the image; The hyperspectral curve RP images of cotton plants in three different states were grayed, and then the gray-level gradient co-occurrence matrix (GLGCM) was used to capture the second-order statistics of grayscale and gradient information in the grayscale recursive map image, and a total of 15 texture features were extracted; Different GLGCM feature parameters are used to characterize the texture distribution of the recursive image from different angles. Small gradient advantage, large gradient advantage, grayscale distribution unevenness, and gradient distribution unevenness describe the distribution of grayscale and gradient in the image texture respectively; the angular second-order moment reflects the uniformity and coarseness of the texture distribution; the grayscale mean and gradient mean characterize the concentration of grayscale and gradient in the image texture; the grayscale mean square error and gradient mean square error reflect the change cycle of the image texture; the correlation describes the similarity of local grayscale and gradient in the image in rows or columns; the grayscale entropy, gradient entropy, and mixed entropy characterize the complexity or non-uniformity of the image texture; the moment of inertia characterizes the grooves and clarity of the image texture; the inverse difference moment reflects the homogeneity and regularity of the image texture; The Pearson correlation analysis was carried out to evaluate the correlation between the texture features and the three-class labels, and the three-class labels of healthy cotton plants, non-symptomatic plants and symptomatic plants were set. The correlation between 15 texture features and the three-class labels was evaluated through Pearson correlation analysis to determine the texture features that were moderately or strongly correlated with the occurrence of cotton Verticillium wilt; the original color recursive image was converted into a grayscale image, and 15 texture features were extracted using the GLGCM method. The absolute values ​​of the correlation coefficients of 11 texture features were higher than 0.6, showing a strong correlation with the class labels of healthy, non-symptomatic and symptomatic Verticillium wilt cotton plants. Among these features, five were related to entropy and asymmetry. For the remaining four texture features, two of the texture gradient features had a moderate correlation with the three-class labels (0.4< |r| < 0.6), while the remaining two features, namely grayscale mean and grayscale asymmetry, were weakly correlated with the class labels (0.2< |r| < 0.4). The four texture features were removed to reduce the complexity and computational cost of the input features; The principal component analysis (PCA) is used to eliminate redundant information of features, and the principal component dimension reduction processing is performed on 11 texture feature parameters. The eigenvalues ​​of the first five principal components are greater than 1, and the cumulative variance contribution rate of these five principal components is as high as 99.02%, which basically explains most of the information of the 11 texture feature parameters. Therefore, the texture feature parameter information of the first five principal components characterizing the recursive graph of the spectrum of cotton verticillium wilt in different states is determined; the three-classification cotton plant verticillium wilt detection model of "healthy-unsymptomatic-symptomatic" is established using the K nearest neighbor method (KNN), support vector machine (SVM), extreme learning machine (ELM) and extreme gradient boosting machine (XGBoost). The confusion matrix of the test set of the four models shows that in the KNN model, 7 healthy samples are misclassified as unsymptomatic, and 8 symptomatic samples are misclassified as unsymptomatic. In the SVM model, 6 healthy samples are misclassified as unsymptomatic, and 5 symptomatic samples are misclassified as unsymptomatic. In the ELM model, 6 healthy samples were misclassified as no symptoms, and 6 symptomatic samples were misclassified as no symptoms. In the XGBoost model, 3 healthy samples were misclassified as no symptoms, and 3 symptomatic samples were misclassified as no symptoms. By comparison, it can be seen that the XGBoost model has the least misclassification, followed by the SVM and ELM models, and the KNN model has the most misclassification; the accuracy, precision, recall, and F1 scores are used to further compare and analyze the performance of different machine learning classification models. Among the four machine learning classification models, the KNN classification model has the lowest overall classification performance. The classification performance of the ELM and SVM classification models are improved compared with the KNN model, and the corresponding classification indicators have increased. The accuracy, precision, recall, and F1 score of the XGBoost classification model are 96.3%, 95.6%, 96%, and 95.8%, respectively, showing the highest classification ability.

Images