复杂环境下口罩佩戴检测算法研究

Research on Mask Wearing Detection Algorithm in Complex Environment

在疫情防控的要求下,口罩佩戴检测受干扰因素、复杂环境的等多方面的影响,高效的佩戴检测具有重要意义。为解决上述问题,提出一种基于YOLOv5的改进算法。首先,在复杂场景下,对口罩数据集进行预处理,其中包括频域图像增强算法及Mosaic、Mix up图像融合算法;其次将YOLOv5的网络模型CSP模块换成CSP1-1;部分CBL模块换成CBS模块;并且加入三个CBAM注意力机制模块,使其在保持原有模型的基础上使其检测精度上提高了1.9%,能够对行人进行实时准确的口罩佩戴检测。

Under the requirements of epidemic prevention and control,mask wearing detection is affected by interference fac-tors,complex environment and other aspects,so efficient wearing detection is of great significance.To solve the above problems,an improved algorithm based on YOLOv5 is proposed.Firstly,mask data set is preprocessed in complex scenes,including frequency domain image enhancement algorithm,Mosaic and Mix up image fusion algorithm.Secondly,change the network model CSP mod-ule of YOLOv5 into CSP1-1.Some CBL modules are replaced with CBS modules.In addition,three CBAM attention mechanism modules were added to improve its detection accuracy by 1.9%on the basis of maintaining the original model,enabling real-time and accurate mask wearing detection for pedestrians.