| As the world’s largest vegetable consumer, the disease stress for vegetables leads to China’s economy and people’s life suffer from a severe losses. This phenomenon has become the main factor which result in huge harm for vegetable production’s efficiency and safety. Meanwhile, grey mould is a kind of serious threat as to vegetable production in the fungus diseases, and it not only seriously occurs in the plants growing period.but also in postpartum storage and transportation procedure,and at the mean time it can continue to cause serious harm.This reserch is mainly focused on the tomato plant, which is been widely planted in our country and also it has high economic value. What’s more, it is a one of the major vegetables for urban and rural residents, by Deuteromycotina Botrytis Cinerea as pathogen which leads to gray mold on vegetalbles. A new method and system were developed for gray mold early diagnosis of tomato leaves and canopy by using hyperspectral imaging detection technology. A new theory and method that based on tomato grey mold early detection was found,which can enforce early, accurate and no-destructive diagnosis. This research aim to improve the plant resistance mechanism research in our country, control the extent of harm which caused by grey mold, and promote the development of plant pathology. The main creative results were achieved as follows:(1) A new segment method for plant and other objects by using plant’s unique red edge phenomenon was put forward. By using to calculate the variance image for the hyperspectral image bands from680-740nm, set a binary image, effectively separatedthe tomato leaves from the white paper, effectively separated the tomato plant from the background of soil, tray and other background information. Meanwhile, a new segment method for plant and other objects using to calculate the variance image for the hyperspectral image bands from1350-1500nm, set a binary image, effectively separated the tomato leaves from white paper, effectively separated the tomato plant from the background of soil,tray and other background information.(2) A spectral analysis technical method was proposed as spectral preprocessing-effective wavelength extraction-effective wavelength images and band combination--calibration model. The early, fast and high precision and nondestructive models for gray mold on tomato leaves based on hyperspectral image were developed. A complete comparison was performed among raw spectra and different spectral preprocessing methods, and Standard Normalize Variate(SNV) data preprocessing was used in visible and short near infrared image (400-1000nm) analysis. Soft independent modeling of class analogy (SIMCA) was applied for effective wavelengths (EWs) selection, in which the discrimination power parameters were used,and extracted401nm,491nm,550nm,625nm649nm,687nm and743nm as feature wavelengths in the visible short wave of near infrared spectral region. In long wave of near infrared spectral region extraction1355nm,1446nm and1608nm as feature wavelengths, and then linear polynomial characteristics based classifier--PLSDA, the statistical parameters based classifer—BayesC and LDC.and adaptive learning based classifier—BPNN and SVC were set up, and find statistical parameters based classifier is the most suitable for this study.(3) Separately set multiple linear regression (MLR) prediction model by using the effective bands in visible and shortwave near-infrared wavelengths image and long wave near-infrared wavelengths image,and thus gain band combination images for early detection for grey mold on tomato leaves.In the visible and shortwave near-infrared spectral bands combine with the variance image of680-740nm and segment image, A complete comparison was performed among five edge detection methods were used,and then choose Laplace sharpen combined with Sobel edge extraction method to remove the shadows information in segment image and extract leaves information. A complete comparison was performed among spatial based sharpen and frequency based sharpen were used, and then choose high frequency emph.In the long wave near-infrared spectral bands combine with the1530nm image and effective combine image, A complete comparison was performed among spatial based sharpen and frequency based sharpen were used, and then choose high frequency emphasize filter as sharpen method to extraction gray mold information, calculate the ill area and area ratio.(4) Research on using MSR light compensation and wavelet fusion algorithm to reduce the canopy tomato image causing by light inhomogenous phenomenon, and through the image and corresponding band statistics, such as variance, information entropy, clarity, and twisted index, correlation coefficient, and deviation index to judge whether fusion image can be characterized the original data information; different input data preprocessing method for tomato grey mold canopy early detection model were studied.and the Normalize preprocessing method was chosed as visible and near infrared hyperspectral image shortwave and long wave near infrared hyperspectral pretreatment method; The vegetation index characteristics extracted image method, PCA wavelet image extraction method and the use of statistical formula extraction method were studied, based on visible in the near infrared shortwave hyperspectral tomato’s canopy grey mold early detection using740-900nm’s variance image as as feature image, and can establish tomato grey mold visualization detection; Based on long wave near infrared hyperspectral tomato’s canopy grey mold early detection, the1300-1450nm’s standard varicance statistical image was used as the feature image, and can establish tomato grey mold visualization detection. |
PDF Full Text Request