基于卷积神经网络的纸张表面缺陷智能检测算法研究
Research on Intelligent Detection Algorithm of Paper Surface Defects Based on Convolutional Neural Network
王娟 1王卫斌 1康晓梅1
作者信息
- 1. 西安翻译学院,陕西 西安,710105
- 折叠
摘要
针对纸张缺陷检测领域中如何有效提升缺陷特征提取能力、提高检测精度以及减少小目标缺陷漏检的难题,创新性地提出了一种基于改进Faster R-CNN算法的检测方法.该方法通过采用ResNet-50 代替传统的VGG16 作为特征提取的骨干网络,有效地增强了对纸张缺陷特征的捕获能力;进一步引入CBAM模块,实现了对空间及通道注意力的双重优化,显著提升了缺陷检测的准确度.此外,通过将ROI-Pooling技术升级为ROI-Align技术,本方法进一步增强了模型对纸张缺陷检测的泛化性能.经验证,该改进算法在常见纸张缺陷检测方面的平均精度达到了 98%,不仅显著提高了检测精度,还有效减少了小目标缺陷的漏检,降低了错误检测率,为纸张缺陷检测技术的发展提供了新的思路和方法.
Abstract
Aiming at the problem of how to effectively improve the ability of defect feature extraction,improve the detection accuracy and reduce the missing detection of small target defects in the field of paper defect detection,this paper innovatively proposes a detection method based on the improved Faster R-CNN algorithm.By using ResNet-50 instead of traditional VGG16 as the backbone network of feature extraction,the method effectively enhances the ability of capturing paper defect features.Furthermore,CBAM module is introduced to realize the dual optimization of space and channel attention,which significantly improves the accuracy of defect detection.In addition,by upgrading ROI-Pooling technology to ROI-Align technology,this method further enhances the generalization performance of the model for paper defect detection.It is proved that the average accuracy of the improved algorithm in the detection of common paper defects reaches 98%,which not only significantly improves the detection accuracy,but also effectively reduces the missed detection of small target defects and reduces the error detection rate,providing a new idea and method for the development of paper defect detection technology.
关键词
卷积神经网络/纸张/缺陷/Faster/R-CNN算法/注意力机制Key words
convolutional neural network/paper/defect/Faster R-CNN algorithm/attention mechanism引用本文复制引用
基金项目
陕西省科技厅自然科学基础研究项目(2022JQ-712)
出版年
2024
NETL
NSTL
万方数据