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

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.