Non-destructive Detection of Citrus Huanglong Disease Using Hyperspectral Image Technique

Abstract:

In order to explore the feasibility of the quick nondestructive detection of citrus Huanglong disease, the hyperspectral image technique combined with least square support vector machine (LS-SVM) and partial least squares discriminate analysis (PLS-DA) were used. The hyperspectral images of the normal, the Huanglong disease of slight, moderate and serious, the lack element citrus leaves were collected in wavelength range of 380~1080nm. By using variance analysis method, the differences in content of chlorophyll, soluble sugar and starch of leaves of the normal, the Huanglong disease of slight, moderate, serious and the lack element were analyzed, and the chlorophyll, soluble sugar and starch were the indicator which could be used to discriminate Huanglong disease. The partial least squares (PLS) method was adopted to establish the mathematical model of quantitative analysis of chlorophyll, soluble sugar and starch, and root mean square error of forecast model were 7.46, 5.51, 5.88 respectively, which provided the basis for rapid detection of citrus Huanglong disease hyperspectral images. The average spectrum of hyperspectral images was extracted in interested area. The differences in absorbance at 750nm was found by analyzing five kinds of leaves of representative spectrum of the normal, the Huanglong disease of slight, moderate and serious, the lack element. The 2order derivative was used to process the sample spectrum, the baseline drift in 450~650nm and 800~1000nm band was eliminated and the effective spectral information was enlarged. Using principal component analysis (PCA) and successive projections algorithm (SPA) to screen the input variables of the model of least squares support vector machine (LS-SVM) qualitative discrimination of citrus Huanglong disease, the LS-SVM model was built for qualitative discrimination and compared with the partial least squares qualitative discriminate model (PLS-DA) at the same time. The prediction sample set which was used to evaluate the performance of model was not used to establish the model. The results showed that the accuracy of PLS-DA model of citrus Huanglong disease was higher, three leaves of lack element were misclassified as serious Huanglong disease, and the misclassification rate was 56%. The experimental results showed that the hyperspectral image technology combined with PLS-DA can achieve rapid and nondestructive detection of citrus Huanglong disease and the degree of Huanglong disease.