为从荧光磁粉探伤图像中获得精确的裂纹提取结果，提出一种基于几何与纹理特征相融合的磁粉探伤裂纹提取算法。算法首先根据裂纹的几何形状特征，提取图像里亮线中心的脊线作为待定裂纹；然后提取待定裂纹上各脊线点的尺度不变特征变换（SIFT）描述子向量，取均值作为待定裂纹的纹理特征，并用支持向量机分类器识别。实验结果表明：针对磁粉探伤中的裂纹，该算法在查全率和查准率上都比传统单一使用几何形状特征或纹理特征的算法更高，结合裂纹的几何形状和局部区域纹理有效地提升了裂纹提取的精度。
This paper introduces an algorithm based on shape and texture features to recognize cracks from fluorescent magnetic particle inspection images. The algorithm detects ridge lines in the images that represent candidate cracks, extracts the local texture features of each ridge point using a scale-invariant feature transform (SIFT) descriptor, and averages those from the same line to form the texture features of the ridge line. A support vector machine classifier is then used to detect the cracks. Tests show that this algorithm more accurately distinguishes the cracks from non-defects than conventional shape-based algorithms or texture-based algorithms. The combined shape and local textures improve the crack detection accuracy.
[1] 赵万民, 曾德文. 荧光磁粉探伤缺陷识别系统的研究现状与关键技术[J]. 铁道机车车辆工人, 2008(8):1-4. ZHAO W M, ZENG D W. Research status and key technologies of defect recognition system for fluorescent magnetic particle inspection[J]. Railway Locomotive and Rolling Stock Workers, 2008(8):1-4. (in Chinese)[2] 康健, 杜向党, 白龙. 火车轮对荧光磁粉探伤图像的裂纹识别技术研究[J]. 机械与电子, 2013(12):44-46.KANG J, DU X D, BAI L. Image recognition technology research of train wheel based on phosphor magnetic powder flaw detection[J]. Machinery and Electronics, 2013(12):44-46. (in Chinese)[3] 李远江, 张华, 王姮, 等. 基于支持向量机的磁粉探伤智能识别算法[J]. 西南科技大学学报, 2015, 30(1):66-70.LI Y J, ZHANG H, WANG H, et al. Intelligent recognition algorithm based on support vector machine magnetic particle detection[J]. Journal of Southwest University of Science and Technology, 2015, 30(1):66-70. (in Chinese)[4] MAHENDRA A, STOLZ C, MERIAUDEAU F, et al. Automated inspection of tubular material based on magnetic particle inspection[C]//BINGHAM P R, LAM E Y. Image processing:Machine vision applications V:Proceedings of SPIE. Bellingham, USA:SPIE, 2012:830007.[5] 张静, 杨薛涛, 王姮, 等. 荧光磁粉探伤智能图像识别技术研究[J]. 西南科技大学学报, 2014, 29(2):82-86.ZHANG J, YANG X T, WANG H, et al. Research of intelligent image recognition technology in fluorescent magnetic particle flaw detection[J]. Journal of Southwest University of Science and Technology, 2014, 29(2):82-86. (in Chinese)[6] 王姮, 卜燕, 张华, 等. 基于分形维数的磁痕图像缺陷检测[J]. 计算机应用研究, 2015, 32(2):603-605.WANG H, BU Y, ZHANG H, et al. Defect detection of magnetic particle indication image based on fractal dimension[J]. Application Research of Computers, 2015, 32(2):603-605. (in Chinese)[7] STEGER C. An unbiased detector of curvilinear structures[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1998, 20(2):113-125.[8] STEGER C. Unbiased extraction of lines with parabolic and Gaussian profiles[J]. Computer Vision and Image Understanding, 2013, 117(2):97-112.[9] LOWE D G. Distinctive image features from scale-invariant keypoints[J]. International Journal of Computer Vision, 2004, 60(2):91-110.[10] 谢凌曦. 基于局部特征的图像表示模型理论与实践[D]. 北京:清华大学, 2015.XIE L X. Image representation models based on local features:Theory and practice[D]. Beijing:Tsinghua University, 2015.[11] CORTES C, VAPNIK V. Support-vector networks[J]. Machine Learning, 1995, 20(3):273-297.[12] MA T, SUN Z G, ZHANG W Z, et al. A machine vision assisted system for fluorescent magnetic particle inspection of railway wheelsets[C]//CHIMENTI D E, BOND L J. AIP conference proceedings:Volume 1706. Melville, USA:AIP Publishing, 2016:150003.
