Abstract:The article proposes an intelligent recognition method for transformer substation straps based on an improved YOLOv5 and DBSCAN clustering approach. The aim is to achieve intelligent monitoring and efficient management of the switch status of straps. To address the small and densely distributed target of strap switches, the original YOLOv5 network is improved by adding detection heads to enhance the ability to detect small targets, and the EIOU loss function is used to optimize the accuracy of target box localization. By combining with the DBSCAN clustering method, the center points of the detected target boxes are clustered, and each strap''s target box is arranged according to their relative positions. By comparing the switch status in the four tables and one guide, intelligent comparison of all strap switches is achieved. This method can improve monitoring accuracy and management efficiency, reduce manual intervention costs, and provide strong support for the safe and stable operation of the power system.