摘要
全路段跟车拍摄隧道漏缆卡扣图像再进行逐张排查,是实现卡扣故障检测的重要手段.针对目前LBP、CS-LBP等相关变体算法存在描述子质量不佳、特征维度过高的问题,提出MD-LBP关联方向特征提取算法实现故障卡扣的检测工作.该算法首先对输入图像进行高斯滤波预处理,根据图像的全局灰度均值得到图像的自适应阈值;其次计算图像的三层MD-LBP特征图结构,依次经过两次下采样分别得到Cell主方向特征和Block主方向特征;然后再在Block特征图上提取关联方向特征,并以此作为描述子;最后通过SVM区分故障卡扣,完成检测工作.实验还对比了LBP、HOG等9种特征描述子的检测情况,结果表明该算法不仅特征维度低,并且检测故障卡扣的召回率和精准度都达到了85%.
Taking images of the leaky cable buckle in whole-section railway tunnels and checking them one by one is an important means to realize the detection of buckle faults. Aiming at the problems of poor descriptor quality and overly high feature dimensions in LBP, CS-LBP and other related variant algorithms, the MD-LBP and correlated directional features extraction algorithm is proposed to realize the detection of buckle with faults. First, the algorithm performs Gaussian filter preprocessing on the input image to obtain the adaptive threshold of the image according to the global gray mean value of the image. Second, it calculates the three-layer MD-LBP feature map structure of the image, and gets the Cell main direction features and Block main direction features after two down samplings. Then, the associated directional features are extracted on the Block feature map and used as descriptors. Finally, the detection is completed by distinguishing the buckle with faults through SVM. The experiment also compares the detection of 9 feature descriptors, such as LBP and HOG. The results show that the MD-LBP not only has low feature dimensionality, but also achieves 85% recall and accuracy in detecting leaky buckle.
作者
张云佐
宋洲臣
杨攀亮
ZHANG Yunzuo;SONG Zhouchen;YANG Panliang(School of Information Science and Technology,Shijiazhuang Tiedao University,Shijiazhuang 050043,China)
出处
《北京交通大学学报》
CAS
CSCD
北大核心
2021年第5期101-107,共7页
JOURNAL OF BEIJING JIAOTONG UNIVERSITY
基金
国家自然科学基金(61702347,61972267,62027801)
河北省自然科学基金(F2017210161)。
关键词
故障检测
漏缆卡扣
关联方向特征
支持向量机
局部二值模式
fault detection
leaky cable buckle
correlated directional features
support vector machine
local binary pattern
