Application of near Infrared Spectroscopy to Detect Mould Contamination in Tobacco

Abstract

Mould infection is a significant postharvest issue for processors of tobacco, which can cause direct product loss and value reduction of product as well as serious economic losses. However, mould mostly is undetectable at the early stages using traditional sorting techniques. In this study, near infrared (NIR) spectroscopy was used to detect mould infection in flue-cured tobacco samples. Based on visual analysis grading a good to bad (GBA) algorithm for feature selection of NIR spectra and linear discriminant analysis routines were applied to detect mould contamination. The optimal wavelengths of NIR spectra included bands at 1066 nm, 1130 nm, 1832 nm and 1474 nm which were applied to establish a classification model which achieved a low classification error rate (2.92% with a Wilks's λ of 0.216 at P < 0.001). The classification accuracies of unmould and mould were 94.6% and 96.9%, respectively (100.0% for slight mould, 100.0% for low mould, 95.0% for medium mould and 92.9% for high mould). A sorting system was developed based on multispectral NIR bands. This showed rapid, accurate and effective detection, and identification of mould-contaminated tobacco was possible at early stages of mould contamination, making it possible to remove mould-contaminated tobacco found in tobacco lots.