Rapid Nondestructive Detection Of Citrus Greening(HLB) Using Hyperspectral Imaging Technology

Citrus greening, also known as dieback, is a major threat to the Gannan navel orange industry. Currently, visual detection based on typical symptoms of HLB and PCR detection is the main methods for screening citrus greening. However, it is time-consuming and expensive. In addition, the visual detection is subjective, easily confused HLB fruit trees with other disease fruit trees have the similar symptoms. In recent years, Hyperspectral imaging technology, as a new generation of spectral analysis technology, while providing spectral information and spatial information of objects simultaneously, has widely be used in nondestructive testing research for the quality of agricultural product. In this paper, use the Gannan navel orange as the research object; propose several ways to detect HLB based on hyperspectral imaging technology. Main research contents and conclusions are as follows:1) Elaborate the mechanism of HLB using hyperspectral detection, based on the difference in the physicochemical index and the difference in the spectral response characteristic. Three physicochemical indexes, include chlorophyll, starch and soluble sugar of each leaf were analyzed chemically. Then, combine these three indexes to draw a three-dimensional map. It can be seen that the three kinds of blades were presented clustering phenomenon in the three-dimensional map. The phenomenon show that the chlorophyll, starch and soluble sugar could be used for the identification of HLB leaves. Subsequently, ANOVA analysis showed that the three kinds of physicochemical indexes have significant difference(P<0.05), and show in the response features in hyperspectral reflectance spectra. After ANOVA analysis, analyses differences between two groups in the four groups with LSD analysis. According to the LSD analysis, these three physicochemical indexes must be combined in the spectroscopic diagnosis of HLB2) Establish the HLB classification model based on hyperspectral imaging system. Collected hyperspectral images of 240 navel orange leaf, including normal leaf, HLB leaves, nutrient deficiency leaves. Extract the region of interest(ROI) according to the ratio of leaf area, and extract the average spectral of ROI to build a Fisher discriminant model. The MD method and the GA algorithm have been used to extract the characteristic bands, and optimization the model. Finally, a cross validation method has been used to validate the reliability of the model. Fisher discriminant model with the bands extracted by MD method have the best result, discriminant accuracy rate of the initial group is 93.8%; for the cross validation group, the identification accuracy rate was 93.3%.3) Establish predict model for chlorophyll, soluble sugar, starch based on hyperspectral imaging system. The SPA algorithm and the CARS algorithm have been used to extract the characteristic bands, greatly reducing the number of variables, thus simplifying the model, and the optimization of the model results. A comparison between SPA and CARS is that CARS selected more characteristic wavelengths, but the model is better. For the calibration set and prediction set in the CARS-PLS model, the correlation coefficient was 0.98, 0.93; 0.88, 0.75; 0.80, 0.81.4) The chlorophyll, soluble sugar, starch were quantified by CARS-PLS model, and used to establish the PLS-DA model of HLB. In the PLS-DA modeling, 103 samples were randomly selected as the training set, and the remaining 35 samples were used as the prediction set, the best number of principal components was 3. For the training set and prediction set, the correlation coefficient was 0.92, 0.91, the discriminant accuracy rate was 91.3%, 88.6%.