采用分类学习的电气化机车受电弓拉弧检测方法

An Arc Detection Approach for Pantograph of Electric Locomotive Using Classification Learning

针对高铁等电气化机车受电弓拉弧实时监控中的高虚警对机车运行带来的干扰问题,提出了一种分类学习的实时拉弧检测和报警方法。该方法首先通过梯度投影快速确定受电弓所在区域,然后在受电弓区域寻找潜在拉弧区域,接着提取区域内边缘梯度直方图,并用分类学习方法判断其是否存在拉弧,最后采用多帧平滑滤波方法确定长持续受电弓拉弧。实验结果表明,采用多帧平滑滤波的方法相对于单帧拉弧检测方法能提升正确检测率约8%,同时降低虚警约32%。

An online arc detection approach using classification learning is proposed to solve the interference problem during locomotive running due to high false alarms in the real-time on-line monitoring of pantograph arc detection for electric locomotives. The approach firstly determines pantograph region through gradient projection. Then candidate arc regions are searched in the determined region, histograms of gradient image in these regions are extracted, and the classification learning approach is used to determine whether a frame is an arc or not. Finally, a multi-frame smoothing based approach is proposed to detect the pantograph arc clip. Experimental results show that the proposed multi-flame smoothing based approach improves the correct detection rate by about 8% and reduces the false alarm by about 32%.