An Improved YOLO Leucocyte Classification and Recognition Method

In order to solve the problem of leukocyte category recognition under complex background, YOLOv3 model is improved in this paper.First, use the K-means algorithm to cluster the anchor boxes to obtain three different sizes of anchor boxes for potential targets to be identified. Then, the squeeze excitation network are introduced into the basic network of the YOLOv3 network to improve the matching Target recognition accuracy and scale utilization degree. The experimental results show that in the 11 types of leukocyte classification tasks, the average average accuracy of the improved YOLO network is 0.937. Compared with the original YOLO network, the improved method has better performance in leukocyte classification and recognition.