Classification and recycling of recyclable garbage based on deep learning

Abstract

Garbage classification and recycling have a lot of benefits, which help protect water and soil resources, improve the living environment's quality, and speed up the green circular economy development. However, traditional garbage collection is weak in effectiveness, requiring a lot of workforces, material, and financial resources. This paper combines ShuffleNet v2 and the depth-separable convolution method to create lightweight YOLOv5s for classifying and positioning recyclable waste. Experimental results show that the enhanced model is only 62% parameters of the original model. In the case of the input resolution being 640 × 640, the mAP (mean Average Precision) of the enhanced model is 94% in accuracy, which is 2.1% higher than the original YOLOv5s. Regarding speed, the reference time is 11.5% faster than the original YOLOv5s on Jetson Nano. In addition, compared with the current mainstream target detection models, the proposed model also expresses the characteristics of recyclable garbage well and can provide a reference value for the classification of recyclable garbage and the lightweight development of recycling.