Abstract

Chestnut is one of the important agricultural products, especially in Asia. However, a large number of chestnuts are lost due to mildew after harvest. A critical method for preventing and reducing large-scale mildew is to identify and remove moldy chestnuts. Additionally, this procedure plays a significant role in the processing of chestnut-based food product. In this work, we identified moldy chestnuts by near infrared spectroscopy. Near infrared spectra for 833 to 2500 nm were acquired from 109 chestnut samples, including 40 chestnuts without mildew, 40 chestnuts with severe mildew and 29 chestnuts with slight mildew and they were used for establishing a discrimination model. A separated set of samples with 3 mixed groups, including chestnuts without mildew (n = 20), chestnuts with severe mildew (n = 20) and chestnuts with slight mildew (n = 8), were used for the validation of the model. The results show that the optimal classification model was achieved based on the spectra band of 1818 - 2085 nm by using the first derivative and vector normalization for spectra preprocessing and the Ward’s algorithm as distance algorithm method. The correct classification rates of sound chestnuts, slightly moldy chestnuts and severely moldy chestnuts were 100, 92.8 and 100% respectively. These results demonstrated that the discrimination model based on near infrared spectral analyses can be used to accurately identify moldy chestnuts. 

Key words: NIR, supervised pattern recognition, nondestructive detection, chestnuts, mildew.