Early detection of mechanical damage in Chinese winter jujube (Zizyphus jujuba Mill. cv. Dongzao) using NIR hyperspectral images

Abstract. Winter jujube (Zizyphus jujuba Mill.) is an important kind of fruit in China. It‘s favored for good taste and abundant nutrition. But its fruit are sensitive and can easily develop brown spots after suffering mechanical stress during mechanical harvesting and postharvest handling, which cannot be detected easily by machine vision in very early stages of maturity. Thus an NIR hyperspectral imaging system was used to detect mechanical damage. Images were captured after the damage was produced by dropping fruit from different height in order to estimate the moment in which the damage could be effectively detected in the images. For reducing the dimensionality of the data, three feature selection methods were used. Results revealed there are two common wavebands No. 130 (1353 nm) and No. 232 (1691 nm) in all the feature selection method. In addition, three classifiers were evaluated to segment the images of the mangos into two classes: damaged and nondamaged. Among these classifiers, SVM offering the best performance which reached 95.16% with the selected features set by the Consistency. Hence, this work lays the foundations for future implementation of a multi spectral on-line platform capable of detecting early damage caused by mechanical stress.