Garbage Sorting System Based on Composite Layer CNN and Multi-Robots

Abstract

Object detection technology is the key problem in the garbage sorting system. There are many difficulties in identifying the target in the garbage, such as poor visibility and targets being coverage. The existing image processing methods are difficult to detect the target objects in the garbage sorting system. To solve this problem, this paper proposes a garbage sorting system based on convolution neural network and robots. The CNN network combines the features of three different network layers and predicts the location and angle information of the target objects at the same time, which we called composite layer. Coupled with the training solution of classification regression and angle prediction, the performance of object detection is further improved. The entire sorting system is composed of three subsystems: vision system, electrical system and mechanical system. The vision system is responsible for the image processing and send the position and attitude information to the electrical system. The electrical system controls the different robots to do sorting motion. We construct a TestSet which has 23983 objects to be detected. The experiment results show that our system's Drop rate is 1.7% and False rate is 4.8%, which is good enough to meet the needs.