Optimization of the YOLOv5 deep learning model for peripheral blood cell detection and recognition
Yunfei Gao, Y.-Y. Liu, Shilei Zhao, Xingxing Ma, Yun Zhou · 2024
In recent years, as neural networks continue to evolve, the use of YOLO deep learning algorithms in medical imaging and diagnosis has become increasingly prevalent. The detection and recognition of blood cells are crucial aspects of medical diagnosis. While deep learning-based object detection and recognition are garnering increasing interest, detecting and counting blood cells in medical imaging remains an essential and challenging task. In this paper, we optimize and improve the YOLOv5 algorithm to achieve efficient detection and recognition of peripheral blood cells. We begin by outlining the architecture and the training procedures of YOLOv5 and discuss its potential applications in medical imaging. Building on the characteristics of blood cell images, we introduce an optimization method using DenseNet based on YOLOv5. Enhancements in data preprocessing, network structuring, and training parameters have improved the accuracy and efficiency of blood cell recognition and counting. Experiments conducted on our proprietary dataset and subsequent comparisons with the base algorithm affirm our claims. Results suggest that our proposed method demonstrates strong performance in detecting and identifying blood cells, offering high practicality for applications.