Study On Diagnosis Of Vegetable Foliage Diseases Based On Computer Vision And Spectral Analysis

Recently, the digital, normalized, quantitative and undestroyed methods tend to be popular in plant disease diagnosis. The symptoms of plant disease are complicated. However, current methods for detecting plant disease are behindhand and inefficient. This study was aimed to develop rapid, nondestructive methods to determine the diseases of cucumber and tomato, and the plant pathogenic pathogens, comprehensively utilizing the technologies of computer vision, digital image processing, pattern recognition, hyperspectral imaging, spectroscopy analysis, chemometrics, plant pathology and plant pathogenic mycology etc.. Based on the above study, rapid detection models and systems for vegetable foliage diseases and pathogens were established.1.A diagnosis system for recognition of 9 tomato foliage diseases was developed, using computer vision technique, digital image processing, and pattern recognition technology. This diagnosis system can automatically identify diseases including tomato early blight (Alternaria solani), tomato late blight (Phytophthora infestans), tomato leaf mould (Fulvia fulva), tomato corynespora leaf spot (Corynespora cassiicola), tomato pseudocercospora leaf spot (Pseudocercospora fuligena), tomato septoria leaf spot (Septoria lycopersici), tomato gray leaf spot (Stemphylium solani), tomato powdery mildew (Sphaerotheca fulifinea) and tomato myrothecium leaf spot (Myrothecium roridum). An image acquisition system was established to acquire leaf images. Tomato diseased leaves were collected from different varieties and ecological regions to establish the digital image database. The image pre-processing techniques were applied to segment the lesion regions from diseased leaves. And then 9 color characteristics, 5 texture characteristics and 4 shape characteristics of the lesion regions were extracted quantitatively. The discriminate models were developed, and the overall classification accuracy achieved 94%. On the basis of the above study, a diagnosis system for recognition of tomato foliage disease was developed on MATLAB platform.2.Hyperspectral imaging (400-720nm) was investigated for the detection of normal and diseased cucumber leaf samples with powdery mildew (Sphaerotheca fuliginea), angular leaf spot (Pseudomopnas syringae), downy mildew (Pseudoperonospora cubensis), and target leaf spot (Corynespora cassiicola). A hyperspectral imaging system was established to acquire and pre-process leaf images, and then the hyperspectral image database of diseased cucumber leaves was established. Stepwise discriminate and canonical discriminate were executed to reduce the numerous spectral information. By stepwise discriminate, 12 optimal wavelengths from the original 55 wavelengths were selected, and the classification accuracies achieved 100% and 94% for the training and testing sets, respectively. By canonical discriminate, the 55 wavelengths were reduced to 2 canonical variables, and the classification accuracies for the training and testing sets were both 100%. These results indicated that it is feasible to identify and classify cucumber diseases using hyperspectral imaging technology and discriminate analysis.3.Fourier transform infrared spectroscopy (FTIR) technique is available for the early detection of corynespora spot on cucumber leaves before the symptoms appeare.Comparing the FTIR spectra of healthy cucumber leaves with the infected cucumber leaves at various times post-infection,three sensitive bands 1735cm^-1,1545cm^-1 and 1240cm^-1 were selected for the identification of cucumber corynespora leaf spot.According to the peak areas at these sensitive bands,diseased and healthy cucumber leaf samples could be classified correctly.Results clearly demonstrated that FTIR is an available technology for the early detection of corynespora spot on cucumber leaves before the symptoms appeare.4.Based on the diversity and specificity of spore morphology, a plant pathogenic fungi automatic identification system was developed. This system could identify Erysiphe, Corynespora, Alternaria, Botrytis, Cercospora, Colletotrichum, and Phyllosticta. A microscope image acquisition system was established to acquire images of fungal spores, and a digital image database was established. The spore images were segmented with mathematical morphology method. Shape features (e.g. perpendicular, circulation and shape similarity energy) and texture feature (e.g. normalized intensity distribution of the spore image regions) were extracted. A Bayes classification algorithm based on the shape information is proposed to classify the spores in the image,and the classification accuracies achieved 89%. Eventually, a diagnosis system for recognition of plant pathogenic fungi was developed on MATLAB platform.5.A method was developed for identification of plant pathogenic fungi from 27 fungal strains, using the Fourier transform infraredspectroscopy (FTIR), chemometrics, and multivariate statistical analysis. Spectra of fungi were acquired with FTIR-ATR method (wavenumber region 4000 to 800 cm^-1), and high-resolution and well-reproducibility infrared spectra were obtained. Significant spectral differences among these strains were observed in the wavenumber regions of 3050-2800cm^-1, 1800^-1485cm^-1, 1485^-1185cm^-1, and 1185-900cm^-1. According to the characteristic bands in these regions, cluster analysis and canonical vitiate analysis were executed to classify the FTIR spectra, and the classification accuracies were above 97%. In conclusion, the FTIR-ATR is a feasible tool for fungal strain identification and classification.This was an interdisciplinary research, which combined the diagnosis theory of plant disease with modern information technology to detect plant diseases and pathogens. On one hand, disease was detected with technologies of computer vision, hyperspectral imaging, and Fourier transform infrared spectroscopy, respectively using the information of image, image and spectra, and infrared spectra. On the other hand, fungal phytopathogens were discriminated with methods of image processing and Fourier transform infrared spectroscopy, according to the morphology and biochemical characteristics of fungi respectively. The dissertation provided a new method for rapid diagnosis of plant diseases and pathogens, and gave an example of modern information technology application in agriculture.