Detection of Helmet Wearing Based on Improved Yolo v3

Aiming at the problem of helmet recognition in complex working environment, an improved Yolo v3 algorithm is proposed. The backbone network of the improved Yolo v3 algorithm is the darknet-53 network. The size is changed by changing the size of the convolution kernel, and the feature selection is performed by using the full convolution method. The k-means++ clustering algorithm is used to optimize the selection method of the anchor box to solve the problem of unstable initial centroid. GIoU is used as the bounding box coordinate regression loss in the design of the loss function, which solves the scale-sensitive problem of mean square error loss. In addition, Focal Loss is introduced to reduce the weight of simple background classes, so that the algorithm model is more focused on target detection. Finally, the algorithm is applied to the helmet wearing detection task. The experimental results show that all the evaluation indexes of the improved algorithm model perform well and meet the accuracy requirements of practical application detection.