基于深度学习的智能垃圾分类系统研究

Research on Intelligent Waste Classification System Based on Deep Learning

目前我国对于智能化垃圾分类回收仍处于完善推广阶段。文中应用重构优化后的YOLOv5神经网络模型,搭配YOLOv5.s权重,TensorRT加速等手段能取得良好的智能垃圾分类识别效果,在其自建数据集准确率达95.23%以上,移动端部署识别速度达120 fps以上,同时搭配多级分类机构和物联网云平台等手段,能较好地实现智能垃圾分类系统社区化部署。有望解决日常生活中生活垃圾的自动分类问题,进一步缓解了日益增加的垃圾种类多、分类困难的问题,促进了垃圾分类的普及。

At present,China is still in the stage of improvement and promotion for intelligent waste classification and recycling.This paper applies the reconstructed and optimized yolov5 neural network model and yolov5.s weight,tensorrt acceleration and other means to achieve good intelligent waste classification and recognition results.The accuracy of its self-built data set is more than 95.23%,and the recognition speed of mobile terminal deployment is more than 120 fps.At the same time,it is combined with multi-level classification mechanism,Internet of things cloud platform and other means.It can better realize the community deployment of intelligent waste classification system.It is expected to solve the problem of automatic classification of domestic waste in daily life,further alleviate the increasing problem of many types of waste and difficult classification,and promote the popularization of waste classification.