Face Detection Based On YOLOv3 With Improved Loss Function

Face detection technology is a pre-step of face recognition technology and has always been valued by people.With the development of deep learning in recent years,face detection technology has also made great progress as an important research direction of computer vision.Face detection is affected by factors such as complex environment,expression,occlusion,posture changes,etc.,and the detection accuracy and speed often fail to reach the expected target.YOLOv3 network is a very good target detection algorithm,it can not only meet the real-time requirements of detection,but also get better detection accuracy.Aiming at the problem of missed detection and misdetection in YOLOv3 detection,this paper proposes a YOLOv3 face detection algorithm based on improved loss function.First,through the K-means clustering algorithm,9 prior frames suitable for face datasets are obtained;then,a new bounding box regression loss function Generalized Intersection over Union and balanced YOLOv3 are proposed The single-step positive and negative sample unbalanced focus loss is used as a confidence function to improve the detection accuracy;finally,the residual block is used to improve the final target detection layer of the network.The improved YOLOv3 model is trained and tested on the Wider Face database.The experimental results show that the face detection method based on the improved loss function YOLOv3 proposed in this paper is better than the original method,and the improvements made are feasible and effective.