Abstract: The proposed YOLOv8-HTPCP detection algorithm, based on an enhanced version of YOLOv8, addresses the challenge of detecting contact points in the arch mesh of high-speed trains. This algorithm incorporates Swim-transformerV2 into the backbone network to enhance model training stability and accuracy, enabling improved information extraction from the entire scene. GSConv is utilized to reduce complexity while maintaining model accuracy. Furthermore, integrating a lightweight Ghost module with the YOLOv8 algorithm significantly reduces network parameters. Ablation and comparison experiments were conducted using an arch net contact point dataset. The results demonstrate that the average accuracy of the YOLOv8-HTPCP model reached 98.5% for mAP_0.5 value, 97.8% for mAP_{0.5 ~ 0.95} value, and 96.7% for reference number 2659823. Compared to the original YOLOv8 model, there was a respective increase of 1.6%, 3.9%, and 1.8% in mAP_0.5 value, mAP_{0.5 ~ 0.95} value ,and recall rate; meanwhile, there was a decrease of 14.7% in parameter count.This study provides valuable technical insights for detecting geometric parameters in bow meshes.