Automatic defect identification and grading system for ‘Jonagold’ apples

Abstract

A method to grade `Jonagold' apples based on features extracted from defects is described. Database consisting of multi-spectral images of Jonagold apples is used for the work. Fuzzy C-Means (FCM) clustering method is used for defect segmentation, features from defect part is extracted using Histogram of Oriented Gradients (HOG) method and Apple classification is performed by using Multi-Class Support Vector Machine (MSVM) with accuracy of 97.5% for two category grading (healthy and defected) and 94.66% for multi-category grading (healthy apples, slightly defected apples and seriously defected apples).