APPLICATION OF A FUZZY CLASSIFICATION TECHNIQUE IN COMPUTER GRADING OF FISH PRODUCTS

This work presents the enhancement and application of a fuzzy classifi cation technique for automated grading of fish products. Common featur es inherent in grading-type data and their specific requirements in pr ocessing for classification are identified, A fuzzy classifier with a four-level hierarchy is developed based on the ''generalized K-nearest neighbor rules.'' Both conventional and fuzzy classifiers are examine d using a realistic set of herring roe data (collected from the fish p rocessing industry) to compare the classification performance in terms of accuracy and computational cost. The classification results show t hat the generalized fuzzy classifier provides the best accuracy at 89% . The grading system can be tuned through two parameters-the threshold of fuzziness and the cost weighting of error types-to achieve higher classification accuracy, An optimization scheme is also incorporated i nto the system for automatic determination of these parameter values w ith respect to a specific optimization function that is based on proce ss conditions, including the product price and labor cost. Since the p rimary common features are accommodated in the classification algorith m, the method presented here provides a general capability for both gr ading and sorting-type problems in food processing.