Application of near infrared spectroscopy to detect mold contamination in tobacco
Mold infection is a significant postharvest issue for processors of tobacco, which can cause value reduction of product and direct product loss. However, mold is mostly undetectable at early stages by traditional sorting techniques. In this paper, Near-Infrared (NIR) spectroscopy technique used in detection of the percentage of mold infection in tobacco samples was studied. A good to bad (GBA) algorithm for feature selection with visual analysis grading Linear Discriminant Analysis (LDA) routines was applied, which achieved low classification error rates as 2.92% of total error, with a Wilk’s λ 0.216 (P < 0.001). The optimal features corresponded to Abs [1066 nm], Abs [1130 nm], Abs [1832 nm], and Abs [1474 nm]. The accurate classification of 5.43% unmold and 3.06% mold (0.00% slight mold, 0.00% low mold, 5.00% medium mold and 7.14% high mold) error was achieved. According to the results, the sorting system developed based on multispectral NIR bands showed the potential for rapid detection and removal of mold contaminated tobacco as well as the reduction of the incidence of early mold contamination in tobacco lots.