铁路车辆监测图像识别模型训练及验证平台研究  

Training and verification platform for railway vehicle monitoring image recognition model

作　　者：杨凯 张淼 祁苗苗 YANG Kai;ZHANG Miao;QI Miaomiao(Institute of Computing Technologies,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China)

机构地区：[1]中国铁道科学研究院集团有限公司电子计算技术研究所,北京100081

出　　处：《铁路计算机应用》2023年第6期26-30,共5页Railway Computer Application

基　　金：中国国家铁路集团有限公司科技研究开发计划项目(N2021J017)。

摘　　要：针对铁路车辆轨边图像检测系统现有图像自动识别模型训练及评价过程中训练数据不足、数据质量不高、评价标准不一致等问题,研究铁路车辆监测图像识别模型训练及验证平台。设计故障图像数据统一接入,专家标定数据形成,自动识别模型接入、训练、对比评测等方法,为故障图像自动识别模型提供标准训练数据、统一评测验证与管理服务的能力。实践表明,该平台实现了车辆故障图像数据的集中汇总与统一管理,为铁路车辆监测图像自动识别技术的发展提供了有力支持。Aiming at the problems of insufficient training data,low data quality and inconsistent evaluation standards in the training and evaluation process of the existing image automatic recognition model of the railway vehicle trackside image detection system,this paper studied the training and verification platform for for railway vehicle monitoring image recognition model,designed methods such as unified access to fault image data,expert calibration data formation,automatic recognition model access,training,comparative evaluation,etc.,provided the ability of standard training data,unified evaluation,verification,and management services for fault image automatic recognition models.The practices show that this platform implements centralized collection and unified management of vehicle fault image data,provides strong support for the development of automatic recognition technology for railway vehicle monitoring images.

关 键 词：轨边图像检测系统 安全监测 图像识别 模型训练 数据标定 

分 类 号：U279[机械工程—车辆工程]