SH-YOLO: Enhanced Real-Time Detection of Laparoscopic Surgical Instruments in Computer-Aided Surgery Based on Star Operation and Hybrid Attention Mechanisms

Abstract

Real-time surgical instrument detection is essential in computer-aided surgery systems for procedure identification, quality maintenance, and operation evaluation. To meet the real-time detection requirements of laparoscopic surgical instruments, a dataset for laparoscopic surgery is established, and an enhanced YOLOv5 algorithm named SH-YOLO is proposed. SH-YOLO incorporates star operation, denoted as element-wise multiplication, for efficient feature extraction and a hybrid attention mechanism (HAM) for adaptive feature fusion, facilitating real-time detection of laparoscopic surgical instruments. Specifically designed for computer-assisted surgical environments, SH-YOLO addresses critical challenges including non-central object localization, subtle feature discrimination, and motion blur resilience. Experimental results show that, compared to YOLOv5, SH-YOLO significantly enhances detection speed and accuracy. Specifically, SH-YOLO achieves a mean Average Precision (mAP) of 91.10%, an F1 Score of 0.941, and a detection speed of 97.4 FPS, outperforming YOLOv5 by 8.77%, 0.03, and 35.1 FPS, respectively. These results validate SH-YOLO’s capability to meet the high accuracy and real-time requirements of instrument detection in computer-assisted laparoscopic procedures, which provides reliable technical support for intelligent surgical navigation and robotic control systems.