Fast Detection Methods For Crop Disease Infection Period Using Spectral And Imaging Technology

As one of the most frontier technologies in modern agriculture, digital agriculture is the key and core technology for the efficient, ecological, safe and sustainable development of modern agriculture. Digital and information technology, which could provide agricultural procedure and management with fast and accurate information acquisition, scientific decision-making and high-efficiency job control, has become a global hotspot in the field of agricultural science and technology. Food safety and supply security are one of the most fundamental purposes of agricultural production. Therefore, how to realize rapid monitoring and precise management during crop growth has become a top priority of the development of agricultural science and technology. Crop growth process is influenced by many factors, and disease is a major factor which could cause yield losses and affect food safety. Thus the prevention and control of crop diseases, especially detection and monitoring in the early time, plays a vital role to prevent diseases from spreading and causing a severe reduction in yield and quality. The main results were achieved as follows:(1) This study focuses on the comprehensive utilization of Raman spectroscopy, spectroscopy and imaging technology and hyperspectral imaging technology to develop rapid detection methods for crop diseases and establish detection models under different infection periods. This study provides a new approach for rapid diagnosis and accurate identification of crop disease during its early infection, which has a vital significance to control and cut off diseases from spreading. Thus it contributes to the implement of precise and digital management of agriculture.(2) The effective recognition models based on Raman spectroscopy was developed for ultra early diagnosis of Phyllosticta theaefolia Hara of tea. The confocal micro Raman spectroscopy response mechanism and characteristics of tea leaf cell wall components infected by Phyllosticta theaefolia Hara was analyzed combined with effective wavelength selection and chemometric methods, The results indicated that principal component analysis could cluster the Raman spectra data of the healthy and diseased samples, providing a qualitative distinction. The effective information extraction method which combined biPLS and SPA improved the efficiency of modeling. Linear formula of classification model based on14effective wavelengths about health and infected tea cell wall could identify healthy and diseased samples. RBF-NN and LS-SVM could accurately identify health and disease tea, with discriminant accuracies of100%.(3) The rapid recognition models were developed for early diagnosis of two tomato diseases, including early blight disease and gray mold disease, and realized the rapid detection of tomato leaf SPAD value under disease stress. Based on spectroscopy and imaging techniques, the early diagnosis and accurate identification of tomato leaf disease using the visible and near infrared wavelengths (400-1000nm), near infrared wavelengths (900-1700nm) and texture feature were achieved combined with effective wavelength selection methods. The results indicated that the correct recognition ratio for the model of early blight disease was over91%, the correct recognition ratio for the model of gray mold disease was100%. The optimal model for predict SPAD value under diseases stress was PLS model based on CARS variable screening method. The coefficient of determination of prediction (R2) was0.866, and the root mean square error of prediction set (RMSEP) was2.328.(4) The rapid detection methods and models were developed for diagnosis of Sclerotinia sclerotiorum of oilseed rape. Hyperspectral imaging was applied, and the spectral data were extracted.13spectral pretreatment methods,3different classification threshold value (0.5,0.3,0.1) of PLS-DA model were compared, and4different discriminant models (PLS-DA, LDA, BPNN, LS-SVM) were compared. The results indicated that the discriminant accuracy rate of BPNN model with five effective wavelengths which were selected by SPA was highest. The discriminant accuracies in the calibration set and the prediction set were93%and89%, respectively. Based on the texture features extracting from the feature images and the effective texture features selected by SPA from the extracted texture features, PLS-DA, LDA, BPNN and LS-SVM models were built and the discriminant results were similar, the discriminant accuracy of the calibration set was above80%.(5) Two sets of plants disease image feature extraction and recognition software system were developed based on VC++and Matlab. The software system could extract characteristic parameters of different crop diseases, calculate disease leaf area ratio, and discriminate infected period and level of a specific disease. It provided a new approach for disease early diagnosis.