Visual detection of sprouting in potatoes using ensemble-based classifier

This paper introduces novel methods for detection of sprouting in potatoes using machine vision. An ensemble-based classifier was trained to detect sprouting considering the diversity of both feature extraction and classifiers. Two categories, namely manual feature extraction and automatic feature extraction, were utilized to enhance the diversity of feature extraction. In experiments, the proposed ensemble-based classifier without multiple channels CNN (MC-CNN) outperformed mainstream methods and achieved state-of-the-art prediction rate of 0.916 and f-measure of 0.905 under lower standard deviation. Furthermore, with the help of three different CNN classifiers, there was already an obvious improvement in the performance of ensemble-based classifier with MC-CNN. The prediction rate and f-measure increased about 4 approximate to 5%, comparing to that without MC-CNN. The results indicate that this approach has a better capability to combine the features for detection of sprouting in potatoes, in which the diversity of both feather extraction and classifier had been enhanced. Practical Applications Discrimination and classification of potato sprouting or not, which uses ensemble-based classifier system with combine the best features.