Apple Stem and Calyx Recognition by Decision Trees

In this paper, a decision tree-based approach for recognizing stem and calyx regions of apples by computer vision is proposed. The method starts with background removal and object segmentation by thresholding. Statistical, textural and shape features are extracted from each segmented object and these features are introduced to two decision tree algorithms: CART and C4.5. Feature selection is accomplished by sequential floating forward selection method. Analysis showed that feature selection improves accuracy of both system. Eventhough CART performed slightly better than C4.5 after feature selection, McNemar’s test found them statistically indifferent.