Phenotypic Monitoring Of Crop Diseases Based On Hyperspectral Image Analysis

Crop disease phenotypic monitoring is of great significance for crop resistance breeding and cultivation management.At present,traditional disease surveys are mainly conducted through artificial fields.This method is time-consuming,labor-intensive,subjective,and inefficient.This study aims at the problems mentioned above,and focuses on the two types of diseases,foliar disease and stalk disease.Taking tea tree anthracnose and rice sheath blight as examples,the spectral response characteristics and differences of these two diseases are studied based on imaging hyperspectral data Based on different types of diseases,different analysis ideas and methods are proposed,and atlas analysis is used to identify diseases and insect pests.The main work of experiments and research in this paper is as follows:(1)Carry out systematic experiments to obtain research data.Hyperspectral data of rice sheath blight and tea tree anthracnose were obtained by designing and conducting experiments.In the field experiment of Sheath Blight of Hangzhou Rice Research Institute,a hyperspectral acquisition experiment was carried out on Sheath Blight imaging.60 shots of imaging spectral data(30 susceptible and 30 normal)were shot.After spectral quality monitoring,32 were selected for experimental data.analysis.We also conducted tea anthracnose experiments in Wuyi County,Zhejiang Province.We collected healthy and anthracnose leaf samples in the experimental tea garden,transported the leaves from the field to the laboratory with a small incubator,and used a set of indoor hyperspectral imaging systems to perform imaging spectral test experiments.100 tea leaves(50 normal and 50 anthracnose leaves).Damaged and abnormally lighted leaves were removed,and 78 leaves(39 normal and 39 anthracnose)were finally selected as samples for subsequent research.These experiments provide a solid foundation for subsequent disease data analysis.(2)Taking tea tree anthracnose as an example,a hyperspectral image recognition method for foliar diseases is proposed.Based on hyperspectral imaging data,feature construction and optimization analysis were performed,and independent t-test analysis and ratio analysis were combined to screen out disease-sensitive bands at 542 nm,686nm,and 754nm;on this basis,ratio and normalized structures were used,respectively.Two kinds of disease indexes were constructed,namely tea anthracnose ratio index(TARI)and tea anthracnose normalized index(TANI);based on hyperspectral features that distinguish sensitive anthracnose from normal leaves,an automatic The recognition algorithm realizes adaptive detection of anthracnose lesions on tea leaves.By verifying the samples,the model's overall recognition accuracy(OAA)of the lesion area image can reach 0.98,and the Kappa coefficient can reach 0.94.Compared with direct pixel-based classification results,this method has higher accuracy and stronger robustness,and can provide an effective means for disease identification and degree analysis.Therefore,the use of hyperspectral imaging technology can achieve automatic and accurate monitoring of tea tree anthracnose.(3)Taking rice sheath blight as an example,a hyperspectral image recognition method for stem-type diseases was proposed.Based on the hyperspectral imaging data,a classification model of lesion identification was established,and the background removal was performed in the two-dimensional space of NIR and RED based on the kmeans mean clustering algorithm.Feature extraction and optimization analysis are performed for abnormal and normal parts,and the JM distance and ratio analysis are combined to finally determine the bands at 666 nm and 494 nm as sensitive bands to distinguish abnormal regions.Based on this,Fisher linear discrimination is used to complete the leaf stem Removal of normal parts such as stalks.By using JM distance and T test to screen the disease-sensitive bands,a disease spot recognition algorithm based on the characteristics of the profile curve of the characteristic image was proposed and verified to realize the adaptive detection of rice sheath blight disease spots.Through the analysis and verification of a large amount of data,the overall accuracy of the model recognition of the blight sample image can reach 98.37% on the pixel scale,and the total recognition accuracy on the plaque scale is 87.76%.In addition,compared with the traditional SVM(Support Vector Machines(SVM))method proposed in this study,using the SVM algorithm to extract lesions results in a mixture of diseased and non-spot abnormal areas,and the disease recognition rate is low..Therefore,the algorithm proposed in this study has high accuracy and robustness,and can provide an effective means for disease identification and degree analysis.