基于TX2的改进的YOLOv5口罩佩戴检测算法

Improved YOLOv5 Mask Wearing Detection Algorithm Based on TX2

佩戴口罩是公共场所防疫的重要措施,如何高效、智能地检测口罩的佩戴具有重要的意义。本文采用了一种基于YOLOv5改进的轻量级网络YOLOv5_SN,将其部署在嵌入式平台TX2上,实现了佩戴口罩的实时检测。该模型使用ShuffleNetV2网络代替YOLOv5的特征提取网络,通过深度卷积操作和通道随机混合策略,减少模型参数和计算的数量,最后缩减特征融合层的卷积核数量,进一步压缩模型。实验结果表明,改进后的网络参数量相比YOLOv5降低了93.7%,模型大小减少了91.2%,而mAP@0.5只降低了3.6%,因此,该算法可方便地部署在嵌入式平台上。

Wearing masks is an important measure for epidemic prevention in public places.How to detect the wearing of masks efficiently and intelligently is of great significance.The paper adopted an improved lightweight network YOLOv5_SN based on the improvement of Yolov5,which was deployed on the embedded platform TX2 to realize the real-time detection of mask wearing.In this model,ShuffleNetV2 network was used to replace the feature extraction network of YOLOv5,and the number of model parameters and calculations was reduced by deep convolution operation and channel random mixing strategy.Finally,the number of convolution kernels in feature fusion layer was reduced to further compress the model.Experimental results show that compared with YOLOv5,the number of network parameters is reduced by 93.7%,the model size is reduced by 91.2%.While mAP@0.5 only decreased by 3.6%.Therefore,the algorithm can be easily deployed on embedded platforms.