雨天工况动车组受电弓安全状态检测技术研究
Research on Safety Inspection Technology of EMU Pantographs in Rainy Days
韩彦青 1裴晓将 2张宗灿 3张瑞芳 1周永康2
作者信息
- 1. 中国铁道科学研究院集团有限公司机车车辆研究所,北京 100081;北京纵横机电科技有限公司,北京 100094
- 2. 北京纵横机电科技有限公司,北京 100094
- 3. 中车长春轨道客车股份有限公司,吉林长春 130062
- 折叠
摘要
受电弓是动车组从接触网获取电源动力的关键部件,其完整性和稳定性对于动车组安全运行至关重要.基于车载视频监控系统采集的图像进行受电弓安全状态检测是目前主流的检测方法,面对雨、雪等恶劣天气,如何更好地处理图像以达到精准检测尤为重要.通过自建受电弓雨天数据集,应用深度学习方法进行图像去雨操作;针对不同类别数据,采用基于生成对抗神经网络(GAN)的数据增强方法,对数据集进行增强;通过对YOLOv5算法进行改造,引入SCCONV结构,改善Backbone区域的特征提取能力,测试模型精度达到98％.该检测流程可在高性能边缘计算模块上达到实时分析效果,为我国铁路车载受电弓视频实时智能分析提供了新的思路.
Abstract
Pantograph is a key component for the EMU to obtain power from the OCS,therefore its integrity and stability are crucial to the safe operation of the EMU.Using images collected by the on-board video monitoring system to inspect the pantograph safety is currently the mainstream inspection method.In the face of severe weather such as rain and snow,how to better process images to achieve accurate inspection is particularly important.The paper introduces deep learning for video deraining of the self-built dataset for rainy-day pantograph status,while data augmentation methods based on the Generative Adversarial Network(GAN)is used to enhance the dataset for different categories.Then based on the modification of the YOLOv5 algorithm and the introduction of the SCCONV structure,the feature extraction capability of the Backbone region is improved with a test model accuracy of 98％.The inspection practice helps realize real-time analysis on high-performance edge computing board,offering new idea for real-time intelligent video analysis of railway pantographs in China.
关键词
受电弓/图像去雨/智能识别/视频监控/YOLOv5算法Key words
pantograph/video deraining/intelligent recognition/video surveillance/YOLOv5 algorithm引用本文复制引用
基金项目
中国国家铁路集团有限公司科技研发计划(K2023J006)
出版年
2024
NETL
NSTL
万方数据