A Quality Grading Approach for ’jonagold’ Apples

In this paper we introduce a quality grading system with perceptron networks for ’Jonagold’ apples. The system is composed of pre-processing, classification, post-processing and decision-taking steps. It is tested by a database of 819 apple images and the resulting images are compared with a reference database which is manually segmented. Performances of 2, 3-class classifiers are compared where 3-class system performed better. An absolute error value is introduced to interpret the classification results at pixel level and the advantage of post-processing is shown.