Apple Stem and Calyx Identification with Machine Vision

Abstract

This paper presents an image analysis technique for the identification of apple stems and calyxes. As apple stem and calyx areas appear as dark patches in images, the analysis is focused on the dark patches of fruit surfaces. The patches are first segmented out by a flooding algorithm. To distinguish stem and calyx areas from patch-like blemishes, the three-dimensional shape of an apple geometric surface is used, which is obtained by using a structured light technique. For each patch, the characteristic features are extracted from both the image under normal diffused light and the image with structured light. With these features, back-propagation neural networks are used to classify each patch as stem/calyx or patch-like blemish, to identify stems and calyxes. The proposed technique was tested with sample apples and an average identification accuracy of 95% was achieved.