Early detection of rice blast (Pyricularia) at seedling stage in Nipponbare rice variety using near-infrared hyper-spectral image

Abstract

Blast rice is the worst biological disaster in rice cultivation. It reduces the yield at least up to 40 to 50% (in the worst period of disease). In this study, the near-infrared hyper-spectral image was investigated to detect blast rice in Nipponbare at seedling stage. Two hundred rice seedlings were segregated into two classes: infected and healthy. All of rice seedlings were scanned with a hyper-spectral imaging system in the NIR (900 to 1700 nm) wavelength range. Principal component analysis (PCA) was performed on the images and the distribution of PCA scores within individual leaf were measured to develop linear discriminant analysis (LDA) models for predicting the infected leaves from healthy leaves. An LDA model classified all the leaves into infected and healthy categories, with an overall accuracy of 92% on validation set. Meanwhile, the classification model base on five selected wavelengths (1188, 1339, 1377, 1432 and 1614 nm) was comparable to that base on the full-spectrum image data.