Genetic Algorithm Based Feature Selection for Detection of Surface Defects on Oranges

Abstract

Using machine vision technology to grade oranges can ensure that only good quality of fruits is to be exported. One of the most prominent issues in the post-harvest processing of orange is the efficient determination of skin defects with the intention of classifying the oranges depending on their external appearance. Color, texture, shape and size are the important grading parameters that dictate the quality and value of many fruit products. The accuracy of the evaluation results is increased by proper combination of different grading parameters. This paper present an efficient orange surface sorting system (normal and defect) based on the color and texture features. As a part of feature selection step this paper presents a wrapper approach with genetic algorithm to search out and identify the informative feature subset for classification and then use the classification accuracy of the neural network classifier to determine the fitness in genetic algorithm. The test results showed that the system could be valuable in categorizing the orange surface with better accuracy rate of 94.3%.