基于神经网络融合模型的铁路接触网异物智能检测

Intelligent detection of foreign objects in railway catenary based on neural network fusion model

针对影响铁路接触网正常运行的异物问题,提出了一种基于神经网络融合模型的铁路接触网异物智能检测模型。以Faster R-CNN框架为基础,增加特征金字塔结构以学习图像不同尺度的特征;针对不同异物类型,将其分为鸟巢和轻质漂浮物,并运用ResNet50和ResNet101作为骨架网络,分别针对具有单一特征的鸟巢和特征复杂多变的轻质漂浮物进行识别;融合2个网络的识别框,得到精确的识别结果。对比实验表明,该模型的检测结果优于常规目标检测方法,可有效降低铁路接触网异物检测的人工成本,为铁路接触网的稳定运营提供了可行的解决方案。

This paper proposed an intelligent detection model for foreign objects in railway catenary based on neural network fusion model,aimed at addressing the issue of foreign objects affecting the normal operation of railway catenary.Based on the Faster R-CNN framework,the paper added a feature pyramid structure to learn features of images at different scales,divided different types of foreign objects into bird nests and lightweight floating objects,and used ResNet50 and ResNet101 as skeleton networks to identify bird nests with a single feature and lightweight floating objects with complex and variable features,respectively,and integrated the recognition boxes of two networks to obtain accurate recognition results.Comparative experiments show that the detection results of this model are superior to conventional object detection methods,which can effectively reduce the labor cost of foreign object detection in railway catenary and provide a feasible solution for the stable operation of railway catenary.