This study designed a classification device that automatically identifies garbage. The classification algorithm is performed through a deep learning Convolutional Neural Network (CNN). After testing, the system has been able to effectively distinguish junk objects. In the identification of three types of garbage objects such as iron, aluminum cans, paper lunch boxes, paper cups and plastic bottles, the system identification rate is currently as high as 96.7%. At the same time, the identification rate of abnormal garbage placed on different types of trash cans reached 93.3%.