基于改进的YOLO V3框架的口罩检测

Mask detection based on improved YOLO V3 framework

戴口罩是阻断疫情传播的手段之一,这使得人脸口罩检测系统成为当下人工智能研究的热点之一.然而,不均匀的环境条件如物体遮挡、光照变化等因素,使口罩检测非常具有挑战性.为解决这些问题,本文采用了一种改进的YOLO V3模型即YOLO-口罩模型来检测复杂环境条件下的人脸.通过LWYS方法的应用、密集的建筑整合、空间金字塔池化和Mish函数激活来改进YOLO V3模型,使改进后的模型AP为99.6%,比YOLO V3提升1.7%,检测时间为52.1 ms.

Wearing a mask is one of the means to block the spread of the epidemic,which makes the face mask detection system become one of the hot spots of artificial intelligence research.However,the uneven environmental conditions,such as object occlusion,illumination changes and other factors,make mask detection very challenging.To solve these problems,we adopt an improved YOLO V3 model,namely YOLO-mask model,to detect faces in complex environment.Through the application of LWYS method,intensive building integration,spatial pyramid pooling and activation of Mish function,YOLO V3 model is improved,and the AP of the improved model is 99.6%,which is 1.7%higher than YOLO V3,and the detection time is 52.1 ms.