摘要
塑料制品的产量和种类的飞速增长,给废杂塑料的回收带来极大的挑战;目前面对恶劣和高强度的工作环境,仍然依靠大量人工分拣,无疑亟待自动化升级;为解决上述问题,提出了一种改进的FoveaBox目标检测算法;针对废杂塑料分选背景复杂的问题,采用ResNeXt-101作为主干网络替代ResNet-50来提高特征提取能力;针对外形差异大的问题,采用带缩放系数的可变形卷积来提高卷积过程的有效感受野;针对目标间彼此遮挡问题,采用带层级控制因子的软化加权锚点机制来提高被遮挡目标的检测精度;结果表明,基于改进FoveaBox的废杂塑料检测算法检测平均精度均值达到85.79%,检测速度为71.4 ms,该检测算法较强的实用性得到验证。
The rapid growth in the output and types of plastic products has brought great challenges to the recycling of waste and miscellaneous plastics.At present,facing the harsh and high-intensity working environment,it still relies on mass manual sorting,it is undoubtedly urgent to upgrade automation.In order to solve above problems,an improved FoveaBox target detection algorithm is proposed.In view of the complicated background for waste plastic sorting,ResNeXt-101 is used as the backbone network to replace ResNet-50 to improve the feature extraction ability.Aiming at the problem of large shape differences,a deformable convolution with a zoom factor is used to improve the effective receptive field in the processing of convolution.Aiming at the mutual occlusion problem between targets,a softening weighted anchor point mechanism with hierarchical control factors is used to improve the detect accuracy of obscured targets.The results show that the mean average precision of the waste plastic detection algorithm based on the improved FoveaBox reaches by 85.79%,and the detection speed is up to 71.4 ms,the detection algorithm is strong practically verifyed.
作者
文生平
陈敬福
冯泽锋
朱珂郁
WEN Shengping;CHEN Jingfu;FENG Zefeng;ZHU Keyu(Guangdong Provincial Key Laboratory of Technique and Equipment for Macromolecular Advanced Manufacturing,South China University of Technology,Guangzhou 510640,China;Ministerial Key Laboratory of Polymer Processing Engineering,South China University of Technology,Guangzhou 510640,China;College of Environmental Science and Engineering,Fujian Normal University,Fuzhou 350007,China)
出处
《计算机测量与控制》
2022年第4期66-71,共6页
Computer Measurement &Control
基金
国家重点研发计划(2019YFC1908201)。
关键词
塑料分选
目标检测
改进FoveaBox
可变形卷积
软化加权锚点
plastic classification
target detection
advanced FoveaBox
deformable convolution
soft-weighted anchor point
