TY - JOUR
T1 - Smooth Nonnegative Matrix Factorization for Defect Detection Using Microwave Nondestructive Testing and Evaluation
AU - Gao, Bin
AU - Zhang, Hong
AU - Woo, Wai Lok
AU - Tian, Gui Yun
AU - Bai, Libing
AU - Yin, Aijun
PY - 2014/4
Y1 - 2014/4
N2 - This paper addresses the interpolation issue of current spectral estimation methods in microwave-based nondestructive testing and evaluation. We developed a spatial-frequency feature extraction algorithm for defect detection with an open-ended waveguide system using smooth Itakura-Saito nonnegative matrix factorization. In addition, the mathematical models of spatial-frequency characteristics for both defects and nondefects areas are derived. The newly developed algorithm has two prominent characteristics, which benefit the detection system. First, it is scale-invariant in the sense that spatial-frequency features that are characterized by large dynamic range of energy can be extracted more efficiently. Second, it imposes smoothness constraint on the solution to enhance the spatial resolution of defect detection. To evaluate the proposed technique, we demonstrate the efficacy of the proposed method by performing extensive experiments on four samples: four defects in an aluminum plate with different depths, a steel plate with 15-mm coating thickness, one tiny defect on steel and one natural defect. Experimental results have unanimously demonstrated the capabilities of the proposed technique in accurately detecting defects, especially for shallow and coated samples with high resolution.
AB - This paper addresses the interpolation issue of current spectral estimation methods in microwave-based nondestructive testing and evaluation. We developed a spatial-frequency feature extraction algorithm for defect detection with an open-ended waveguide system using smooth Itakura-Saito nonnegative matrix factorization. In addition, the mathematical models of spatial-frequency characteristics for both defects and nondefects areas are derived. The newly developed algorithm has two prominent characteristics, which benefit the detection system. First, it is scale-invariant in the sense that spatial-frequency features that are characterized by large dynamic range of energy can be extracted more efficiently. Second, it imposes smoothness constraint on the solution to enhance the spatial resolution of defect detection. To evaluate the proposed technique, we demonstrate the efficacy of the proposed method by performing extensive experiments on four samples: four defects in an aluminum plate with different depths, a steel plate with 15-mm coating thickness, one tiny defect on steel and one natural defect. Experimental results have unanimously demonstrated the capabilities of the proposed technique in accurately detecting defects, especially for shallow and coated samples with high resolution.
KW - Defect detection
KW - nondestructive testing and evaluation (NDT&E)
KW - nonnegative matrix factorization
KW - open-ended waveguide
KW - smoothness
KW - source separation
U2 - 10.1109/TIM.2013.2287126
DO - 10.1109/TIM.2013.2287126
M3 - Article
VL - 63
SP - 923
EP - 934
JO - IEEE Transactions on Instrumentation and Measurement
JF - IEEE Transactions on Instrumentation and Measurement
SN - 0018-9456
IS - 4
ER -