Ghost-YOLO: 轻量化口罩人脸检测算法

在嵌入式设备上，由于算力及存储空间的限制，当前的大型高精度目标检测模型的推理速度较低。为此，本文设计了一种轻量化目标检测模型，用于口罩人脸检测。首先，本文设计了一种高激活性鬼影（High Active Ghost，HAG）模块，以轻量的计算代价减少特征图中的冗余。其次，利用HAG实现高激活性鬼影跨段部分（High Active Ghost Cross Stage Partial，HAG-CSP）连接模块，提升了跨段部分连接网络结构的特征学习能力。再次，利用HAG-CSP对你只需看一次（You Only Look Once，YOLO）模型进行轻量化改造来得到完整的Ghost-YOLO网络，并构造出一个口罩人脸检测器。实验结果表明，本文提出方法在NVIDIA Jetson NX嵌入式设备上，在检测精度优于其他目标检测算法的前提下，对于640×640的图片，实现了24.72 ms每帧的检测速度，并且减少了模型的参数量。 

Ghost-YOLO: Lightweight Masked Face Detection Algorithm

? ?In embedded devices， due to the limitation of low computing power and small storage space， the inference speed of current large-scale high-precision object detection algorithms is low. For that， this paper designed a lightweight masked face detection algorithm. Firstly， a high active ghost （HAG） module was designed to reduce the redundancy in feature learning with a light computational cost. Secondly， based on HAG， a high active ghost cross stage partial （HAG-CSP） connection module was designed to improve the feature learning ability. Thirdly， the HAG-CSP was used to lightweight the you only look once （YOLO） model to raise a Ghost-YOLO model for masked face detection. Experimental results show that the speed of proposed method is 24.72 milliseconds per image for 640×640 sized images on a NVIDIA Jetson NX embedded device and reduces the number of parameters， under the condition that the proposed method had a better accuracy performance than existing object detection models.