A Study of Garbage Classification with Convolutional Neural Networks
Shanshan Meng, Wei-Ta Chu
Abstract
Recycling is already a significant work for all countries. Among the work needed for recycling, garbage classification is the most fundamental step to enable cost-efficient recycling. In this paper, we attempt to identify single garbage object in images and classify it into one of the recycling categories. We study several approaches and provide comprehensive evaluation. The models we used include support vector machines (SVM) with HOG features, simple convolutional neural network (CNN), and CNN with residual blocks. According to the evaluation results, we conclude that simple CNN networks with or without residual blocks show promising performances. Thanks to deep learning techniques, the garbage classification problem for the target database can be effectively solved.