Surgical Instrument Recognition Based on Improved YOLOv5

Abstract

Recognition of surgical instruments is a key part of the post-operative check and inspection of surgical instrument packaging. However, manual inventorying is prone to counting errors. The achievement of automated surgical instrument identification holds the potential to significantly mitigate the occurrence of medical accidents and reduce labor costs. In this paper, an improved You Only Look Once version 5 (YOLOv5) algorithm is proposed for the recognition of surgical instruments. Firstly, the squeeze-and-excitation (SE) attention module is added to the backbone to improve the feature extraction. Secondly, the loss function of YOLOv5 is improved with more global parameters to increase the convergence rate of the loss curve. Finally, an efficient convolution algorithm is added to the C3 module in the head to reduce computational complexity and memory usage. The experimental results show that our algorithm outperforms the original YOLOv5 with improvements observed across various metrics: mean average precision 50–95 (mAP50-95) achieved 88.7%, which improved by 1.8%, and computational requirements reduced by 39%. This study, with a simple but effective method, is expected to be a guide for automatically detecting, classifying, and sorting surgical instruments.