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 IEO H 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。%Aiming at automatic identification of rail fastener defects , an image classification and recognition method based on image fusion feature and Bayesian compressive sensing (BCS) was presented .First ,when the fastener sub‐imagewassegmentedfrom thetrackimage ,theimprovededgeorientationhistograms(IEOH)and macroscopic local binary pattern(MSLBP) were extracted separately .Then ,the two were merged into a more robust IEOH_MSLBP fusion feature via hierarchical weighted ensemble . Finally , on the basis of the BCS model , the sensing matrix was constructed based on all training samples , to solve the sparse coefficient vectors corresponding to the test samples , and then to perform the fastener recognition according to the L2 norm corre‐sponding to each of training types in the sensing matrix . The experimental results indicated that the average recognition rate based on the IEOH_MSLBP feature was 5.1% higher than that of the IEOH feature and 4.7%higher than that of the MSLBP feature .Meanwhile ,the proposed algorithm has a higher recognition rate than several mainstream methods , w hich can provide important technical reference for automated inspection of rail‐way fasteners .
展开▼
