Abstract

Info

Author(s):: TIAN Changrui1; LIAO Wei1; XU Zhen2; 1. School of Electronic and Electrical Engineering, Shanghai University of Engineering Science, Shanghai 201620, China 2. School of Mechanical and Automotive Engineering, Shanghai University of Engineering Science, Shanghai 201620, China

Keywords:: Keywords: medical device YOLOv5s attention mechanism α-DIOU deep learning

Abstract:: Abstract: A medical device detection method based on improved YOLOv5s was proposed to solve the problem of medical device stacking and further improve the detection accuracy of medical devices. The proposed method uses C2f module to optimize YOLOv5s network for improving the detection accuracy, introduces squeeze-and-excitation network into the feature fusion network for improving the models attention to effective information, and constructs α-DIOU by introducing Alpha intersection union ratio (α?OU) on the basis of the distance-intersection over union (DIOU) loss function, which makes the bounding box regression more accurate and enables the accurate detection of medical devices in the image. Experimental results show that the improved model has precision, recall rate and mean average precision of 81.8%, 93.7% and 91.5%, respectively, for medical device detection on the validation set, which are 3.2%, 3.4% and 4.6% higher than YOLOv5s model. The proposed method is simple and effective, and is expected to provide new ideas for the detection methods of medical devices.