Open-Set Bottle Classifying using a Convolution Neural Network

Abstract

A multi-class image classification application plays a vital role in our lives. Traditional approaches focus on a close-set classification problem. However, an open-set classification problem often occur in the real-world applications. This paper focuses on the convolution neural network based image classification for beverage bottle image classification under the open-set environment, in which the input image may not appear in any known classes during training time. The proposed models explore the approaches based on the N-Binary, N+unknown, and N+combination models. The results show that N+unknown approach perform better than that of the N+combination and N-Binary approach in terms of accuracy and time.