基于深度学习的轨道交通弓网电弧检测方法  

Arc Detection Method for Rail Transit Bow-Net System Based on Deep Learning

作　　者：郭德龙 晏月 刘虎 Guo Delong;Yan Yue;Liu Hu

机构地区：[1]上海地铁维护保障有限公司 [2]上海应用技术大学

出　　处：《隧道与轨道交通》2024年第3期38-40,75,共4页Tunnel and Rail Transit

摘　　要：轨道交通弓网系统是保证列车平稳运行的关键。弓网电弧会严重影响列车获取电能,对弓网系统造成很大危害。为及时检测电弧的发生,提出了一种基于深度学习的轨道交通弓网电弧检测方法。采用改进的Inception V3网络作为主干网络,以更快、更准地提取电弧特征,选用引导锚框以适应电弧检测中存在的电弧尺寸差异性大的问题。结果表明,基于深度学习方法的轨道交通弓网电弧检测模型的均值平均精度为94.57%,处理速度为0.28 s/幅,远优于其他模型。The rail transit bow-net system is crucial for stable train operation.Arcs in the bow-net system can severely affect the train's power supply and cause significant harm to the system.To detect arcs in real-time,an arc detection method for railway bow-net systems is proposed which is based on deep learning.An improved Inception V3 network is used as the backbone network to extract arc features more quickly and accurately.Guided anchor boxes are selected to address the issue of significant size variations in arcs.The results show that the mean average precision of the proposed rail transit bow-net arc detection model based on deep learning is 94.57%,with a processing speed of 0.28 seconds per frame,significantly outperforming other models.

关 键 词：轨道交通 弓网电弧检测 深度学习 Inception V3网络 引导锚框 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程] U239.5[交通运输工程—道路与铁道工程]