基于计算机视觉的轨道扣件缺陷自动检测算法研究

发布时间：2018-04-26 04:12

本文选题：计算机视觉 + 扣件检测　；参考：《铁道学报》2016年08期

【摘要】：针对轨道扣件缺陷自动识别问题,提出一种基于图像融合特征和贝叶斯压缩感知的图像分类识别方法。从轨道图像中分割出扣件子图像,并分别提取其改进的边缘梯度特征IEOH和扣件端部的宏观纹理特征MSLBP;利用层级加权融合将二者融合形成鲁棒的、分辨力更强的IEOH_MSLBP特征;在贝叶斯压缩感知模型的基础上根据训练样本构建传感矩阵,求解待测试样本相应的稀疏系数矢量,并根据系数矢量中对应训练样本类别的各元素的L2范数判定待测试扣件的状态。实验结果表明,使用IEOH_MSLBP特征的平均检测准确率比单独使用IEOH、MSLBP特征分别高出5.1%、4.7%;同时,本文所提检测方法相较于主流方法在识别率方面具有一定优势,可为铁路扣件自动化巡检提供重要技术借鉴。
[Abstract]:An image classification and recognition method based on image fusion features and Bayesian compression perception is proposed for automatic recognition of track fastener defects. The subimage of fastener is segmented from the track image, and the improved edge gradient feature IEOH and the macro texture feature of the end of the fastener are extracted, respectively, and the hierarchical weighted fusion is used to fuse the two images into a robust and more discriminant IEOH_MSLBP feature. Based on the Bayesian compressed perception model, the sensor matrix is constructed according to the training sample, and the sparse coefficient vector of the sample under test is solved. The state of the fastener to be tested is determined according to the L2 norm of each element corresponding to the training sample class in the coefficient vector. The experimental results show that the average detection accuracy of using IEOH_MSLBP features is 5.1 / 4.7 higher than that of IEOH_MSLBP / MSLBP features alone. At the same time, the detection methods proposed in this paper have some advantages over the mainstream methods in the recognition rate. It can provide important technical reference for automatic inspection of railway fastener.
【作者单位】：西南交通大学机械工程学院;
【基金】：国家自然科学基金(51275431) 四川省科技支撑计划(2012GZ0102,2014GZ0005) 2014年西南交通大学博士创新基金
【分类号】：TP391.41

【相似文献】