CLASSIFICATION OF SWEET ONIONS BASED ON INTERNAL DEFECTS USING IMAGE PROCESSING AND NEURAL NETWORK TECHNIQUES

Maintaining product quality is the key to success in the fresh fruit and vegetable market. Quality assessment with computer vision techniques is possible; however, two basic issues need to be addressed before an automatic sorting system can be developed: (1) which image features best correlate with the product quality, and (2) which classifier should be used for optimal classification. To address these issues, sweet onions were line–scanned for internal defects using x–ray imaging. Spatial and transform features were evaluated for their contributions to product classification based on internal defects. The Bayesian method was used for selecting the salient features. Spatial edge features combined with selected discrete cosine transform (DCT) coefficients proved to be good indicators of internal defects. A neural classifier performed better than the Bayesian classifier for sorting onions into two classes (good or defective) by achieving an overall accuracy of 90%. Losses and false positives were limited to 6% and 10%, respectively. The accuracy, losses, and false positives for the Bayesian classifier were 80%, 16%, and 17%, respectively.