Study On Fast Detection Of Gray Mold On Plants Using Spectral And Multi-Source Spectrum Imaging Technology

The plant disease not only causes the crops underproduction and the quality drops, even has no harvest, but also leads to fungicide unnecessarily increase, which increases the agricultural production cost and leads to the serious environment problem. The plant disease is the one of important factors of the restriction high production, high quality and high benefit agriculture sustained development, therefore it is especially important to monitor to the plant disease, which provides the basis of grasping plant disease's occurrence situation timely, formulating prevention strategy and guiding prevention. This article took solanaceae vegetables such as eggplants and tomatoes and so on as the object of the study, digitizing outward appearance information and the internal chemistry information of the solanaceae vegetables'leaves using hyperspectral imaging detection technology. Related mathematical models were established combining imagery processing technology, the data digging technology and the plant pathology knowledge. The main research conclusions were as follows:(1) The study of early detection of gray mold based on the eggplant leaves using hyperspectral imaging detection technology. This study introduced the preparation process of eggplant leaves sample and the collecting method of hyperspectral imaging information. Using the principal components analyzes (PCA) to reduce the dimensionality of hyperspectral images, which could select 554.05nm, 684.49nm and 763.63nm as the characteristic wavelengths and draw eight textural features parameters based on the grey paragenesis matrix under those three characteristic wavelengths. Successive projections algorithm (SPA) was applied for thirteen characteristic variable selection. Partial least squares (PLS), back propagation artificial neural network (BPANN) and least squares-support vector machine (LS-SVM) models were established separately. The result indicated that LS-SVM model's forecast effect was the best, regardless of the threshold value was set as±0.5,±0.2 or±0.1, the accuracy rate was 97.5%.(2) The study of early detection of gray mold based on the tomato leaves using hyperspectral imaging detection technology. A complete comparison was performed among raw spectra and five spectral preprocessing methods. SPA was applied for spectral feature and image feature of tomato leaves hyperspectral images selection. PLS, BPANN and LS-SVM was established based on spectral information, image information or both spectral information and image information separately. The result indicated that when the threshold value was set as±0.5, PLS, BPANN and LS-SVM acquired satisfied results of early detection of gray mold based on tomato leaves. However as the enhancement of accuracy requirement or the lessening of threshold absolute value, only three models based on both spectral information and image information acquired satisfied results, which reflect hyperspectral images can detect inside and outside of tomato leaves completely. Regardless of the threshold value was set as±0.5,±0.2 or±0.1, LS-SVM model's forecast effect was the best, whose accuracy rate was 100%. (3) Determining the chemical value of eggplant leaves used hyperspectral imaging technology. In order to use hyperspectral information fully and establish more reliable accurate models, SPA was applied for spectral feature and image feature of eggplant leaves hyperspectral images selection. PLS, BPANN and LS-SVM was established separately. The result indicated that BPANN could forecast superoxide dismutase (SOD) of eggplant leaves well and the result was Rp=0.8468. LS-SVM could forecast peroxidase (POD) of eggplant leaves well and the result was Rp=0.8479.(4) Distinguishing different disease of eggplant leaves used hyperspectral imaging technology. SPA was applied for spectral feature and image feature of eggplant leaves hyperspectral images selection. PLS, BPANN and LS-SVM was established to distinguish healthy, gray mold, sclerotinia and continous epidemics of eggplant leaves. The result indicated that BPANN model's forecast effect was the best. When threshold value was set as±0.5, the predicting accuracy rate was 95%. When threshold value was set as±0.2, the-predicting accuracy rate was 85%. When threshold value was set as±0.1, the predicting accuracy rate was 75%.