Non-destructive rapid quality control method for tobacco grading using visible near-infrared hyperspectral imaging

Abstract

The main goal of this study was to investigate the potential of hyperspectral imaging as a non-destructive rapid quality control method for grading cured tobacco bales. Cultivated tobacco plants are harvested and cured before using in the manufacture of tobacco products. Cured tobacco leaf bales are then graded by humans based on visual, physical and sensory characteristics. Hyperspectral imaging can be used as a tool to streamline the tobacco grading process and assist the human graders. Hyperspectral images of cured tobacco bales were acquired using a visible near-infrared (VNIR) hyperspectral pushbroom imaging system (400-1000 nm). Multivariate calibration models were built using end-member extraction and linear discriminant analysis (LDA). The LDA model using Mahalanobis distance metric showed clear discrimination between the different tobacco grades. The relative classification accuracy of this method for Flue Cured and Burley tobacco grades were 93%. This study demonstrates that hyperspectral imaging can be used as a reliable, rapid, non-destructive quality control method for grading cured tobacco bales. Industrial Relevance: This hyperspectral grading system has been developed for tobacco, but can be potentially practiced with other agricultural products such as tea, coffee, grapes etc.