基于深度学习的口罩佩戴识别技术研究

Research on Mask Wearing Recognition Technology Based on Deep Learning

摘要: 本文基于目标检测中口罩佩戴识别的相关研究，针对人群环境下口罩佩戴检测时中小尺寸目标识别等问题，通过在原网络中用稠密模块代替原有的残差模块以及优化损失函数对YOLOv3模型进行改进。实验结果表明，改进后的算法识别平均精度达到95.35%，相较于原YOLOv3算法提升了4.49%，同时该算法还具有较高的检测速度，能够满足实时检测的需求，在实际应用中可以将其嵌入移动式设备，能够有效解放人力，减少管理成本。

Abstract: This paper is based on relevant research in mask-wearing detection for object recognition. Ad-dressing issues such as small-sized object identification in crowded settings, it improves the YOLOv3 model by replacing the original residual modules with dense modules and optimizing the loss function. Experimental results demonstrate that the improved algorithm achieves an average precision of 95.35% in mask-wearing detection, exhibiting a 4.49% enhancement compared to the original YOLOv3 algorithm. Furthermore, this algorithm maintains a high detection speed, catering to real-time detection requirements. It can be embedded in mobile devices for practical application, effectively liberating human resources and reducing management costs.