Identifying defects in images of rotating apples

Abstract

An experimental machine vision system was used to identify surface defects on apples, including bruises. Images were captured through two optical filters at 740 and 950 nm, respectively. In the ensuing grey scale images, defects appeared as dark areas, however, so did shadows and parts of the stem/calyx area. This paper reports a novel approach to locate the defects and eliminate other dark areas. The method is based on rotating the apples in front of the camera while multiple images are acquired. Dark areas, which are found at the same position, relative to the apple, during the rotation, represent defects, while other dark areas, which change shape and/or position from one frame to the next, are not classified as defects. In a test using 54 Pink Lady apples with 56 defects, the system successfully detected 52, or 92% while providing two false positive. In another test with Ginger Gold Apples, where the rotation technique was combined with images of the stem and calyx regions, 90% of the defects were detected with no false positives.