Quality prediction of air-cured cigar tobacco leaf using region-based neural networks combined with visible and near-infrared hyperspectral imaging.

Visible and Near-infrared hyperspectral imaging (VNIR-HSI) combined with machine learning has shown its effectiveness in various detection applications. Specifically, the quality of cigar tobacco leaves undergoes subtle changes due to environmental differences during the air-curing phase. This study aims to evaluate the feasibility of deep learning methods in overcoming data limitations to develop a VNIR-HSI prediction model for the quality of cigar tobacco leaves at different air-curing levels. The moisture, chlorophyll, total nitrogen, and total sugar content in cigar tobacco leaves were predicted across various air-curing stages and light conditions. Results showed that the Diversified Region-based Convolutional Neural Network (DR-CNN) achieved the best performance, with a root mean square error of prediction for moisture at 3.109%, chlorophyll at 0.883 mg/g, total nitrogen at 0.153 mg/g, and total sugar at 0.138 mg/g. Compared to Partial Least Squares Regression and Convolutional Neural Networks, DR-CNN demonstrated superior predictive accuracy, making it a promising model for quality prediction in cigar tobacco leaves during air-curing process. Overall, VNIR-HSI based on DR-CNN can effectively predict the quality of cigar tobacco leaves at different air-curing levels.