A trainable grading system for tobacco leaves

A grading system based on image processing techniques was developed for automatic inspection and grading of flue-cured tobacco leaves. The system used machine vision in extraction and analysis of color, size, shape, surface texture and vein. A two-dimensional feature space was proposed to express feature distribution of tobacco leaves. The space was found to be well confined in an elliptic region. A database was constructed to record the feature distribution of standard contrast tobacco leaves prepared by experts through visual evaluation. The decision on grades was made based on the so-called 'nearest-neighbor' method for which the overall difference among features between the measured tobacco leaves and the standard contrast samples were used as a target parameter for judgment. This system can be easily trained by users with the knowledge of the feature distribution information of different tobacco leaves. It can also be adapted for the inspection of other agricultural products.