Research on YOLOv4 Object Detection Based on K-means Algorithm and Fusion Attention Mechanism

Object detection technology is a machine learning application based on deep learning and computer vision technology, which can replace error-prone manual inspections in various scenarios, such as mask recognition applications. In view of the problems of easy omission and difficult error detection in traditional object detection algorithms, this paper proposes a research on YOLOv4 object detection based on K-means and fusion attention mechanism, with mask recognition as the research object. Firstly, the preprocessing of the dataset is carried out. The K-means algorithm is used to cluster and analyze the face mask dataset, generate more suitable anchor box prior boxes, and use the Mosaic method for data augmentation. Then, the SE module and SPP module of the channel attention mechanism are introduced to improve the basic YOLOv4 model, and the SE-SPP3-YOLOv4 detection model is constructed. The experiment shows that the proposed improved model has significantly improved the detection rate and accuracy of both face and mask targets, and can perform accurate recognition of the wearing of masks on faces in specific application scenarios.