基于深度学习的纸病检测系统设计与研究  

Design and Research of Deep Learning-based Paper Defect Detection System

作　　者：顾文君 谭永涛 李强 刘耀斌 周易 王平军 孙霞[1,3] 陆文荣 吴昱昊 伍沐原 GU Wenjun;TAN Yongtao;LI Qiang;LIU Yaobin;ZHOU Yi;WANG Pingjun;SUN Xia;LU Wenrong;WU Yuhao;WU Muyuan(Jiaxing Vocational and Technical College,Jiaxing,Zhejiang Province,314036;Minfeng Special Paper Co.,Ltd.,Jiaxing,Zhejiang Province,314000;Jiaxing Key Lab of Industrial Internet Security,Jiaxing,Zhejiang Province,314036;Zhejiang Paper Industry Association,Hangzhou,Zhejiang Province,310000)

机构地区：[1]嘉兴职业技术学院,浙江嘉兴314036 [2]民丰特种纸股份有限公司,浙江嘉兴314000 [3]嘉兴市工业互联网安全重点实验室,浙江嘉兴314036 [4]浙江省造纸行业协会,浙江杭州310000

出　　处：《中国造纸》2024年第8期154-159,共6页China Pulp & Paper

基　　金：浙江省高等学校国内访问工程师“校企合作项目”(FG2023285);浙江省教育厅一般项目(Y202351406);嘉兴市应用性基础研究项目(2023AY11022,2024AD10063)。

摘　　要：本课题设计了基于深度学习的纸病检测系统,用于提高造纸生产过程中的质量控制水平。该系统采用了“CCD+FPGA+工业控制计算机+训练计算机”的架构模式,实现了对纸张图像数据的实时采集、纸病的实时判断和纸病类型的实时识别。综合考虑分类准确率与推理速度,选择MobileNet模型算法,其分类准确率达99.5%,每秒可推理约103.1张分辨率为224×224的图像,满足现场纸病图像分类识别的实时要求。A deep learning-based paper defect detection system was designed in this paper to enhance the quality control of papermaking pro⁃duction.This system adopted the architecture model of“CCD+FPGA+industrial control computer+training computer”,achieving realtime collection of paper image data,real-time assessment of paper defects,and real-time identification of types of paper defects.Consider⁃ing both classification accuracy and inference speed,the MobileNet model was chosen to achieve a classification accuracy of 99.5%.It could infer approximately 103.1 images per second with a resolution of 224×224,meeting the real-time requirements for on-site and recogni⁃tion of pager defect image classification.

关 键 词：纸病检测 深度学习 系统设计 架构设计 

分 类 号：TS77[轻工技术与工程—制浆造纸工程] TP391.41[自动化与计算机技术—计算机应用技术]