基于Transformer的二阶段弓网燃弧视觉检测与分析

Two-stage Visual Detection and Analysis of Pantograph-catenary Arc Based on Transformer

为分析地铁运行区段的弓网燃弧发生情况,提出一种基于二阶段的弓网燃弧视觉检测方法,并设计了燃弧视觉检测的后分析流程。一阶段基于改进DETR模型进行弓网接触区域检测,并构建了二阶段弓网燃弧图像分割数据集。接着,采用改进Segformer模型进行二阶段弓网燃弧的图像分割。依据分割结果设计一种燃弧能量等级的匹配方法,并对燃弧发生位置进行定位,最后通过地铁运行实例分析出该区段燃弧发生情况。结果表明,该方法在弓网燃弧检测与分析方面具有较高的准确性和鲁棒性。一阶段弓网接触区域检测模型的mAP_(50)评价指标最高可达89.9%,二阶段弓网燃弧图像分割模型的mIoU评价指标最高可达85.5%。此外,弓网燃弧的分析结果也可为受电弓碳滑板烧蚀损伤的分析提供基础数据。

In order to analyze the arc occurrence of the pantograph–catenary in the subway operation section,a two-stage visual detection method of arch net arc is proposed,and the post-analysis flow of the visual detection is designed.In the first stage,the improved DETR model is used to detect the contact area of the pantograph catenary,and the data set of the arc image segmentation of the pantograph–catenary is constructed.Then,the improved Segformer model is used to segment the two-stage arc image.According to the segmentation results,a matching method of the arc energy level is designed,and the location of the arc occurrence is located.Finally,the arc occurrence situation in this section is analyzed through the example of subway operation.The processing results show that this method has high accuracy and robustness in arc detection and analysis.The mAP_(50) evaluation index of the first stage pantograph catenary contact area detection model is up to 89.9%,and the mIoU evaluation index of the second stage pantograph–catenary arc image segmentation model is up to 85.5%.In addition,the analysis results of the burning arc of the pantograph net can also provide basic data for the analysis of the ablation damage of the carbon slide of the pantograph.