Deep representation for classification of refrigerator image via novel convolutional neural network

Machine vision has played a vital role in household appliance assembly lines in recent decades. Being a practical manufacturing issue, the automatic identification of refrigerators from their front-view images captured from actual scenes is a potentially invaluable tool for industrial production automation. And a large number of approaches have been presented to classify the refrigerators according to their appearance. However, it remains a challenge since there are several hardships in this process. To bridge this gap, we propose an unsupervised convolutional neural network-based pipeline for recognizing the category of each input refrigerator image. By using multi-channel network architecture and a double convolutional operator, the proposed approach can utilize the information from intra-class and inter-class images, simultaneously. To evaluate the performance of this proposed method, we conducted comparison experiments between the state-of-the-art techniques and ours on 39,782 manually sampled images of refrigerators divided into 47 categories. The experimental results demonstrate that the proposed approach outperforms the methods presented in the literature with an accuracy of 99.97%.