Abstract
In the early stage of fruit rot in satsuma mandarins, rot may be difficult for an inspector to detect. We have developed a system for detecting fruit rot by statistical methods and spectral imaging. Rotting satsuma mandarins were prepared by inoculating rotted pieces of another fruit into test samples. Spectra were measured using the device described in Part 1. We developed a model that can discriminate between rotted and normal parts by multivariate analysis using SIMCA, PLS, and MLR. Before analysis, we tested the pretreated fruits using MSC or 2nd derivative methods. The best results were obtained using a combination of 2nd derivative and PLS methods. Spectral images segmented into 5×5 pixel clusters, calculated using the models and representing only the rotted part, were used to judge whether a fruit was rotted. As a result, we were able to detect sections of rot as small as 1 cm in diameter in validation samples.
