Research on Object Detection of Suspension Clamps Based on Attention Mechanism

Aiming at the problems of UAV aerial images, the complex background of dangling wire clip detection can easily cause false detection and missed detection, this paper proposes a new UAV-based dangling wire clip charged detection model-YOLOV8_SL. Backbone's SPPF module is optimised to be SPPF-LSKA, which enhances the network generalisation ability and improves the model's feature extraction ability. Compared with the benchmark model, although the number of parameters of the model and the resources required for its computation are slightly increased, the mAP50 is improved by 2.5 percentage points on the dangling wire clip dataset, and the FPS is 176 for many practical application scenarios, which is able to satisfy certain real-time requirements, and at the same time, there is a better improvement on the problems of false detection and missed detection.