基于深度学习的烟支外观检测系统的设计与实现

Design and Implementation of Cigarette Appearance Inspection System Based on Deep Learning

卷烟机作为卷烟生产企业核心设备之一,具有高速、高自动化的特点,过滤嘴香烟的生产速率可达20 000支/分钟。高速生产下,依靠人工辨别很难发现外观质量不合格的烟支。目前的烟支外观检测系统采购成本高、算法复杂、参数多、参数间具有严格的相关性、调整难度大。因此针对烟棒自动化生产过程中出现的烟支外观缺陷分辨准确度低的问题,设计实现了基于Resnet50的烟支外观检测系统,通过去噪、分割、缺陷特征提取等对烟支图像的缺陷进行分类检测。实验说明,本文提出的缺陷检测算法具有较高的准确率。

As one of the core equipment of cigarette manufacturers,cigarette machinehas the characteristics of high speed and high automation.The production rate of filter tip cigarettes can reach 20000 cigarettes/minute.Under high-speed production,it is difficult to find unqualified cigarettes by manual identification.The current cigarette appearance inspection system has high procurement cost,complex algorithm,many parameters,strict correlation between parameters,and great difficulty in adjustment.Therefore,in order to solve the problem of low resolution accuracy of cigarette appearance defects in the automatic production process of cigarette sticks,a cigarette appearance detection system based on Resnet50 is designed and implemented.The defects in cigarette images are classified and detected through noise removal,segmentation,defect feature extraction,etc.Experiments show that the proposed defect detection algorithm has a high accuracy.