Improved Immunohistochemistry Active Cell Counting Method for YOLOv5s

Abstract

This article proposes an improved YOLOv5s counting method to address the problems of long-term manual counting of positive cells in immunohistochemical images and low consistency. First, by introducing the Triplet attention module, the model focuses on the positive cell area, reducing background interference and improving the network's ability to extract positive cell features; then, a small target detection layer is added to better utilize the semantic information of the network to improve positive cells. recognition accuracy; then, the lightweight up-sampling operator CARAFE is used to improve the quality and accuracy of up-sampling; finally, the WIoU loss function is used to replace the original GIoU of YOLOv5 to enhance model detection performance. Experimental results show that the improved model has an average accuracy of 88.4%, which is 3.1% higher than the original YOLOv5 network model. It can count positive cells quickly and accurately, reducing the workload of doctors.