基于改进VGG-16深度学习网络的防护面罩佩戴识别
Protective Mask Wearing Recognition Based on Improved VGG-16 Deep Learning Network
陈威 1张皓亮 1高崇阳1
作者信息
- 1. 中石化南京工程有限公司, 江苏南京 210000
- 折叠
摘要
为高效识别打磨焊接作业人员是否佩戴防护面罩,提出了改进VGG-16 的深度学习模型,构建了基于VGG-16 的深度特征提取网络挖掘图像的重要信息.为解决VGG-16 网络对图像局部特征和全局结构信息捕捉的不足,建立基于坐标注意力的空间位置信息感知机制,增强对图像位置和通道信息的关注.最后,建立基于多层全连接层的分类网络输出识别结果.实验表明,该模型对打磨焊接作业人员是否佩戴防护面罩的识别准确率、精确率、召回率和 F1 分数分别达到95.88%、96.48%、95.25%和 95.86%,具有比传统人工巡检方法更好的效果.
Abstract
To efficiently identify whether polishing and welding operators are wearing protective masks,an improved deep learning model of VGG-16 network was proposed,and a deep feature extraction network based on VGG-16 was constructed to mine important information of images.To address the shortcomings of the VGG-16 network in capturing local image features and global structural information,a spatial position information perception mechanism based on coordi-nate attention was established to enhance the attention to image position and channel information.Finally,a classification network based on multiple fully connect-ed layers was established to output recognition re-sults.The experimental results showed that the recog-nition accuracy,precision,recall,and F1 score of this model for whether polishing and welding opera-tors wore protective masks reached 95.88%,96.48%,95.25%,and 95.86%,respectively,which had bet-ter performance than traditional manual inspection methods.
关键词
打磨焊接作业/防护面罩/坐标注意力机制/VGG-16网络/深度学习/卷积神经网络(CNN)/智能识别Key words
grinding and welding work/protective face mask/coordinate attention mechanism/VGG-16 network/deep learning/convolutional neural network(CNN)/intelligent recognition引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据