基于YOLO V3神经网络的轨道车辆识别技术研究

Preliminary Study on Railway Vehicle Recognition Technology Based on YOLO V3 Neural Network

传统铁路货车图像识别以车辆号码特征识别为主,存在着受现场环境光照干扰大、安装不便等不足之处,对铁路车辆的识别率低,存在一定安全隐患。针对这一问题,本文利用近年来发展迅速的卷积神经网络、YOLO V3模型、深度学习技术,建立了铁路车辆识别模型,优化了YOLO V3识别算法,利用GPU加速技术减少了CPU的工作负荷,实现了铁路车辆高速通过环境下的车型识别与统计。搭建了简易的试验系统,通过采集图像、数据分析、特征提取等完成铁路车辆车型的识别,在多种环境下进行了试验,并与其它深度学习算法进行了比对,证明设计的有效性及先进性。为进一步提高铁路货运保障技术提供技术铺垫,可以应用在铁路生产、运营、维保等诸多方面,助力铁路货车智能化发展。

Traditional railway freight car image recognition mainly focuses on vehicle number feature,but it has some shortcomings,such as great interference by on-site environmental lighting,inconvenient installation,low recognition rate of railway vehicles,and some hidden dangers.This paper establishes a railway vehicle recognition model using the rapidly developed convolutional neural network,YOLO V3 model and deep learning technology,optimizes the YOLO V3 recognition algorithm,uses GPU acceleration technology to reduce the workload of CPU and realizes the vehicle type identification and statistics under the condition of railway vehicle travelling at highspeed.A simple experimental system was built to complete the identification of railway vehicle models through image collection,data analysis and feature extraction.Experiments were carried out in a variety of environments and compared with other deep learning algorithms to prove the effectiveness and advancement of the design.It provides technical foundation for further improvement of railway freight transport guarantee technology,which can be applied in railway production,operation,maintenance and many other aspects to help the intelligent development of railway freight cars.