Research on laparoscopic surgical instrument detection technology based on multi-attention-enhanced feature pyramid network

Laparoscopic surgery is a very active area of research in clinical medicine. Detection tools surgical videos can help physicians operate instruments, reduce complications, and ensure patient safety. However, the size laparoscopic instruments highly variable, leading to poor detection. Feature pyramid networks (FPNs) effectively solve problem multi-scale target detection, but FPNs still have some problems that limit full utilization features. By analyzing FPN design problem, we propose Multi-Attention Augmented Pyramid Network (MAFPN), which fully utilize First, replace convolutional block with feature selection module (FSM) combines channel attention global attention, selectively maintains important information enhances expressiveness features at each scale. Second, contextual captured by self-attentive augmented fusion (AAFM), enriches high-level effect. Finally, use Dynamic Convolution Decomposition (DCD) alleviate impact upsampling while enhancing expression ability. The experimental results on instrument detection dataset m2cai16-tool-locations show average precision MAFPN 96.5 when IOU 0.5, 1.8% better than baseline method RetinaNet, more 1.6% comparison network. Compared state-of-the-art method, performance superior.