Research on Hidden Danger Target Detection of Transmission Line Shock Hammer under Improved YOLOv8 Algorithm

During the examination of electrical power systems, the prevailing approach involving human-based, hands-on inspections is beset with several drawbacks, including inadequate levels of automation, suboptimal efficiency and inherent safety risks. The unmanned aerial vehicle (UAV) target detection method has broad application prospects for the important link of hid-den danger detection of shockproof hammers. Although the traditional shockproof hammer image detection has limitations such as environmental conditions, field of vision and system performance, we can use a new SR-YOLOV8 Convolutional neural network to achieve efficient and accurate target recognition in complex environments. This network adopts SimAM (Simple, Parameter Free Attention Module), which can effectively reduce the burden on the system and achieve more efficient target recognition. Using Res2Net as the foundation, we can greatly improve the feature extraction effect of the shock hammer feature map. This mechanism is capable of assigning pertinent 3D attention weights to the feature map in accordance with its characteristic attributes, thereby significantly enhancing the precision of the feature map, simultaneously, it broadens the potential applicability of the model. At the same time, a new image dataset for power grid shock hammer detection was created using real detection scenarios, and Mosaic data augmentation strategy was introduced for network training to improve the model's generalization ability and resistance to overfitting. After experimental verification, the improved shockproof hammer detection technology of YOLOv8s has performed well in performance indicators such as accuracy, recall, detection speed, and average accuracy, far exceeding traditional target detection methods, providing a reliable reference for power system inspection.