This paper presents the classification of Jonagold apple defects on their shape, colour and texture basis according to the E.U. standards. The small defects (less than 11 mm2) with little variability were graded separately, with 5 parameters (roundness, mean colour and mean red standard deviation) into three classes (healthy, russeting, defects). With linear discriminant analysis, the global correct classification rate reached 71%. 95% of the healthy blobs, coming from slight over-segmentation, were correctly graded. For larger patches, a first grading between healthy patches and defects (including calyx and stem ends) was carried out. 86% of the patches were correctly graded with 12 parameters: 4 for the shape, 3 for the colour and 5 for the texture. The defects were graded afterwards with the same descriptors set. The results were quite good as a whole (66% of the defects correctly graded). Scabs, stem end and russeting were well recognised, but slight defects and defects rejecting the fruits were less accurately sorted. The use of neural networks enhanced this result (71% of the defects correctly graded).
