钢轨扣件检测技术研究

Rail Fastener Inspection Technology

轨道交通的飞速发展对轨道线路各组成部分的可靠性检测提出了更高要求,依靠人工手持设备检测存在某些弊端,因此迫切需要依靠计算机图像处理和深度学习等技术实现轨道线路的自动化检测。介绍一种基于图像处理的钢轨扣件状态分类方法,该方法通过对扣件图像的二值化处理,根据其水平方向和垂直方向上的投影特点对扣件部分进行分割,随后提取其局部二值模式(LBP)特征并用支持向量机(SVM)进行分类。同时,探讨一种基于深度学习的扣件图像分类识别方法,采用卷积神经网络VGG16架构,该方法中扣件区域的特征提取可由网络自动完成,不仅节省了时间,而且识别精度和适用性也得到明显提升。

The rapid development of rail transit has put forward higher requirements for the reliability test of various track components. The manual handheld inspection devices have some drawbacks. Therefore, it is urgent to rely on computer image processing and deep learning to realize automatic inspection of tracks. A classification method of rail fastener based on image processing is introduced. The method uses binary processing of fastener images to segment the fastener part according to its horizontal and vertical projection characteristics, then extracts the local binary pattern (LBP) feature and classifies it with support vector machine (SVM). At the same time, a deep learning based fastener image classification and recognition method is discussed. The convolution neural network VGG16 architecture is adopted. The feature extraction of fastener area can be automatically completed by the network, which not only saves time, but also improves the recognition accuracy and applicability.