Deep learning technology is widely applied in the field of healthcare

but research on postoperative management of surgical instruments is very limited.To complete the inventory management of surgical instruments

a surgical tool collection system is designed and a commonly used surgical tool dataset(STD)is constructed.An improved network based on the YOLOv5 model is proposed

which fuses a 3D attention mechanism and can extract more comprehensive local and global features.Two attention mechanism embedding strategies:additional strategy and embedded strategy are designed in the experiment.It is verified that the performance of the embedded attention model is superior to that of the additional attention model.It is verified in self-made dataset containing 20 categories that the average accuracy of the improved network is 97.2%.This system provides an effective method for surgical instrument management and has a driving effect on the development of healthcare.