PREDICTING RIPENING STAGES OF BANANAS (MUSA CAVENDISH) BY COMPUTER VISION

Abstract

A computer vision system was implemented to predict the ripening stages of bananas. Our objective was to develop a computer vision algorithm to predict the seven ripening stages based on previously graded bananas by expert visual inspection. Two simple colour features from each image (mean value and variance of the intensity histogram of image) were extracted and analyzed using the RGB, HSV and CIELAB colour spaces with classification purposes. The classification performance of three colour sets of features were compared using discriminant analysis as selection criteria: (1) using colour data extracted from full image of the bananas; (2) colour data from the background of the bananas free of spots (BFS); and (3) combination of the colour data extracted separately from the BFS and brown spots of the banana. Results show that the three evaluated sets were able to correctly predict with more than 94% of accuracy the ripening stages of bananas. The inclusion of colour features from images of brown spots improves the resolution of the classification performance, in particular for stages 4 and 5. In addition, this technique can be extended to evaluate the different quality classes according to the standards proposed by the Codex Alimentaria. Computer vision shows promise as a non-destructive method for on-line prediction of ripening stages of bananas.