Optical Defect Analysis of Florida Citrus

Abstract

Image data from an AT-bus color frame grabber system were stored as HSI (hue-saturation-intensity) components to ascertain types of defects and the ability to discern such defects from color standards. Classification models were evaluated using commercial neural network software and single feature parametric and nonparametric Bayesian classification. The major defect encountered was windscar: 32.5% (grapefruit), 28.5% (orange), and 23.0% (tangerine). Successful classification ranged from 59.3 to 74.2% with neural net models and from 70.2 to 85.8% with single feature Bayesian approaches.