YM_SSH:一种用于实时疲劳检测的轻量级目标检测网络

YM_SSH:a lightweight target detection network for real-time fatigue detection

文章通过轻量化手段实现了在资源受限设备上疲劳检测算法性能提升。同时,为进一步优化疲劳检测精度,本研究在YOLOv3和MobileViT模型的基础上,融入改进的SSH人脸检测网络,将大卷积核替换为若干个3×3卷积核,提出新型轻量化目标检测模型YM_SSH(YOLOv3-MobileViT-SSH)。实验表明,YM_SSH相比于传统轻量级模型(如MobileNetV2-YOLOv3),模型体积缩小了约8 MB,检测精度几乎保持不变,表明YM_SSH模型参数量小,计算成本低,且能获得比其他流行的目标检测模型更高的识别精度和更强的抗干扰能力。

In this study,fatigue detection algorithm performance enhancement on resource-constrained devices is achieved through lightweight means.Meanwhile,in order to further optimize the fatigue detection accuracy,this study incorporates the improved SSH face detection network on the basis of YOLOv3 and MobileViT models,replaces the large convolutional kernel with a number of 3×3 convolutional kernels,and proposes the novel lightweight target detection model YM_SSH(YOLOv3-MobileViT-SSH).Experiments show that YM_SSH reduces the model size by about 8 MB compared to the traditional lightweight model(e.g.MobileNetV2-YOLOv3),and the detection accuracy remains almost the same,indicating that the YM_SSH model has a small number of parameters,low computational cost,and can obtain higher recognition accuracy and stronger anti-jamming ability than other popular target detection models.