White blood cell detection based on improved YOLOv5s

Abstract

In order to effectively solve the problem of classification and identification of human peripheral blood leukocytes, this paper proposes a classification and identification framework for leukocytes based on an improved YOLOv5 network. YOLOv5s is selected as the network model, and the samples are preprocessed by constructing a priori frame through k-means clustering. Secondly, the original feature map is obtained by the Focus slicing operation, and the original feature map is sent to the Neck network, and the information is transferred and fused through FPN and PAN to obtain the optimal weight. Finally, CIOU_Loss is selected as the loss function of frame regression to achieve high-precision positioning. The experimental result shows the average accuracy of white blood cell detection of the improved YOLOv5s algorithm is 94.8%, which is 10.7% higher than the previous one. It shows that this method has high detection accuracy and can effectively improve the accuracy of human peripheral blood leukocyte identification.