|
二维码(扫一下试试看!) |
改进YOLOv7的纸张表面缺陷检测研究 |
Study on Surface Defects of Paper Based on YOLOv7 |
投稿时间:2023-04-28 |
DOI:10.11980/j.issn.0254-508X.2023.11.021 |
关键词: 机器视觉 纸张表面缺陷 纸病 YOLOv7 注意力机制 |
Key Words:machine vision paper surface defects paper defects YOLOv7 attention mechanism |
基金项目:四川省科技计划重点研发项目(2021YFG0055);企业信息化与物联网测控技术四川省高校重点实验室(2022WZJ01);四川省自贡市科技局重点科技计划项目(2019YYJC15);四川轻化工大学科研基金项目(2020RC32)。 |
|
摘要点击次数: 1061 |
全文下载次数: 1033 |
摘要:提出一种改进YOLOv7的纸张表面缺陷一步式检测算法。首先将注意力机制模块CBAM融合到主干和特征提取网络结构,从空间和通道2个维度提取信息,提升小目标纸病特征提取准确性和算法稳定性;将ASPP空洞卷积加入主干网络SPP中,ASPP可以进一步扩大感受野,使较小目标的特征信息在网络传递时得到保留,解决了小目标信息量不足的问题,进而提高小目标纸病识别的性能。通过自制纸病数据集检测实验,与YOLOv7相比,精确率、召回率及平均精确率均值mAP 0.5分别提升了1.5、2.3和2.1个百分点。 |
Abstract:A one-step paper surface defect detection algorithm with improved YOLOv7 was proposed in this study. The attention mechanism module CBAM was integrated into the backbone and feature extraction network structure to extract information from 2 dimensions including space and channel, to improve the feature extraction accuracy and algorithm stability of small target paper disease. ASPP was added to the backbone network SPP, and could further expand the receptive field, so that the feature information of the smaller target was preserved in the network transmission, which solved the small target information insufficient problem, to improve the performance of paper disease recognition for small targets. Through the homemade paper disease dataset detection experiments, compared with YOLOv7, the precision rate, recall rate and average precision rate mean mAP 0.5 of improved YOLOv7 were improved by 1.5, 2.3 and 2.1 percentage points, respectively. |
查看全文 查看/发表评论 下载PDF阅读器 |