国产欧美精品一区二区,中文字幕专区在线亚洲,国产精品美女网站在线观看,艾秋果冻传媒2021精品,在线免费一区二区,久久久久久青草大香综合精品,日韩美aaa特级毛片,欧美成人精品午夜免费影视

Under the requirements of the normalization of the prevention and control of the covid-19 pandemic, the current mask-wearing detection device is affected by numerous factors, such as a large number of people in complex scenes, the easy obstruction of the gathering crowd, and the small size of the inspection target, which are prone to false detections and missing inspections. To solve the above problems, a mask-wearing detection algorithm based on YOLOv5 is proposed to realize real-time detection in complex scenes. Firstly, do Mosaic data enhancement and other processes on the data set. Then, apply the Focus process to remain more complete down sampling information of the images for subsequent feature extraction. Later, Feature enhancement using SPP fusion of multi-scale information and retain spatial information within Neck. Finally, considering the overlapping area, center point distance and aspect ratio between target frame and detection frame, CIoU Loss function is selected s to improve the positioning accuracy. And during the training process, a dynamic adjustment strategy is adopted for the learning rate. Experimental results show that the average accuracy of the improved algorithm reaches 99.3%.