摘要
为解决垃圾分类困难问题,根据大量垃圾图片的特点,提出了基于yolov4的垃圾检测的方法,实现垃圾的自动识别分类和定位。该方法通过手机拍摄以及百度等收集不同种类的垃圾图像集,运用图像增强的方式进行图像预处理,再利用迁移学习模型和k-means++聚类算法进行垃圾网络分类模型的训练测试,最后将训练好的模型通过摄像头对采集的垃圾图片实时识别分类和定位。实验表明:该方法的mAP达到95.15%,能够快速有效地对垃圾分类和定位。
In order to solve the difficult problem of garbage classification,according to the characteristics of a large number of garbage pictures,a method based on yolov4 garbage detection is proposed to realize the automatic identification and classification and positioning of garbage.This method collects different kinds of junk image sets through mobile phone shooting and Baidu,uses image enhancement to pre-process images,uses migration learning model and k-means-clustering algorithm to carry out the training test of garbage network classification model,and finally uses the trained model to identify and classify and locate the collected garbage images in real time through the camera.Experiments show that the mAP of this method reaches 95.15%,which can classify and locate garbage quickly and effectively.
作者
王明吉
陈秋梦
任福深
胡庆
刘博
王一安
WANG Mingji;CHEN Qiumeng;REN Fushen;HU Qing;LIU Bo;WANG Yian(School of Physics and Electronic Engineering,Northeast Petroleum University,Heilongjiang Daqing 163318,China;School of Mechanical Engineering,Northeast Petroleum University,Heilongjiang Daqing 163318,China)
出处
《工业仪表与自动化装置》
2021年第5期20-23,72,共5页
Industrial Instrumentation & Automation
