Analysis of Hyperspectral Characteristics of Coat-oriented Cigar Tobacco with Different Maturity and Construction of a Classification Model

To obtain hyperspectral characteristics of different-maturity coat-oriented cigar tobacco, in this study we used a portable hyperspectral camera to collect hyperspectral images of coat-oriented cigar tobacco at three-maturity stages (SS, CS and WS) and extracted their spectral data. The characteristics of hyper spectrum was analyzed by using multivariate statistical analysis, and a supported vector machine (SVM) model was constructed for discriminating the maturity of coat-oriented cigar tobacco. There existed strong correlation for the bands in the region of 400-680 nm, 730-1000 nm as well as between 400-430 nm and 730-1000 nm, while the correlation between 460-730 nm and 760-940 nm was weak. Three principal components (PC1, PC2 and PC3) with eigenvalues larger than 1 almost contained all the spectral information, and analysis of variance from principal components indicated that there was significant difference among the three-maturity coat-oriented cigar tobacco in visible light, red edge and near infrared region. For SVM, the accuracy of discriminating cigar maturity was more than 98%, and the precision, recall and F1 scores were 100% for CS cigar as well as more than 94% for SS and WS cigar. The hyper spectrum of cigar had multicollinearity with a good dimension reduction effect. The spectral reflection characteristics of different-maturity coat-oriented cigar tobacco was significantly different. SVM had good performance in discriminating maturity of coat-oriented cigar tobacco, which could be used in ripening harvest of coat-oriented cigar tobacco.