Research On Structured Sparse Pattern Decomposition For Nondestructive Testing Technology Of Infrared Thermography

Non-destructive testing is a comprehensive detecting technology that measures and evaluates the physical properties of the surface or inside including various defects and material parameters of test specimens without damaging and destroying the material under test.Optical pulsed thermography is widely used for defect detection of composite materials due to its advantages such as fast detection speed and large detection area.Defects cannot be directly observed from the original image collected by the OPT system due to the influence of noise,therefore,feature extraction algorithms are required to enhance the defect detection.At present,algorithms such as principal component analysis have been widely used in thermal processing,although the existing feature extraction algorithms can enhance defects detections,they have limits or poor performance for detecting weak defects and deep defects.Therefore,more effective algorithm is needed to improve the defect detection rate and resolution.We propose a structured sparse matrix decomposition algorithm based on OPT system to detect defects of composite materials.The proposed algorithm decomposes the given data into three components to enhance the defect detection.The main work of this paper are:1.The thermal image of physical analysis is mainly composed of three components including background,defect and noise by analyzing the thermal image of the specimens with defects.For weak signal extraction problem such as defect detection,most of the observed signals are background traces while only a small amount of defect information exists in the thermographic sequence.The noise is caused by the uneven illumination and the surface emissivity of the specimen.The thermal data collected by the infrared camera can be constructed into a matrix.According to the temperature characteristics,the original matrix can be decomposed into a low-rank matrix,a sparse matrix,and a noise matrix.The low-rank matrix represents the background;the sparse matrix represents the defect.2.The observed signals are not the expected feature signals as they are superimposed by the expected features due to the effects of lateral and longitudinal thermal diffusion.According to the above physical characteristic,the sparse matrix is further factorized into a dictionary matrix and a coefficient matrix.The dictionary matrix represents the thermal features,the weights stored in the coefficient matrix represent the proportion of each thermal feature in a single pixel.The model estimates each component in an iterative and alternating manner.3.In order to estimate the detection ability of different defect detection algorithms,the event-based F-score,SNR and computational time are used to measure the results.Among them,F-Score is used to evaluate the defect detection rate,SNR is used to evaluate the quality of the result image,the computational time is used to evaluate the complexity of each algorithm.From the experimental comparative analysis of multiple test specimens,it can be obtained that the proposed algorithm performs better in detecting weak defects and irregularly shaped specimens compared to other algorithms and significantly improves the contrast ratio between the defective regions and the non-defective regions.The FScore reaches 94%,the SNR reaches 21.91 d B.The F-Score increased by 7%,and the SNR increased by 12 d B compared with the state of the art defect detection algorithms.The computational time of the proposed algorithm is second to the most efficient algorithm,it is acceptable in terms of the computational time for a task whose detection accuracy requirement is higher than the detection efficiency.