混叠场景下的轻量级口罩佩戴检测算法

Lightweight mask wearing detection algorithm in aliasing scenario

针对混叠场景下口罩佩戴检测识别率低,而现有检测模型结构复杂难以部署的难题,提出了一种轻量级口罩佩戴检测算法。首先,轻量化网络MobileNetv3作为混叠场景图像的特征提取网络;其次提出通道混洗,空间上利用不同感受野的卷积核进行特征提取的注意力机制,实现特征信息的强化;最后设计了损失函数解决了数据类不平衡问题,提高了模型检测精度。在公开数据集测试表明,模型平均检测精度为78.1%,FPS达到65.53 Hz,满足在小型设备部署的要求。

A lightweight mask wearing detection algorithm is proposed to address the low recognition rate of mask wearing de⁃tection in mixed scenes and the difficulty of deploying existing detection models with complex structures.Firstly,the lightweight network MobileNetv3 serves as a feature extraction network for aliasing scene images;Secondly,channel mixing is proposed,which utilizes convolutional kernels from different receptive fields in space for feature extraction and enhances feature information;Finally,a loss function was designed to solve the problem of imbalanced data classes and improve the accuracy of model detection.In public dataset testing,the average detection accuracy of the model is 78.1%,and the FPS reaches 65.53 Hz,meeting the require⁃ments for deployment on small devices.