Sparse Reconstruction-Based Thermal Imaging for Defect Detection

This paper proposes an idea of employing sparse reconstruction-based technique for thermal imaging defect detection. The implementation of the reconstruction technique is tested on a carbon fiber reinforced polymer test piece with artificially drilled defects and the test results are compared with the established cross correlation method. The two processes are compared in terms of defect detectability, their SNR variation with defect depth and their computation complexity. When compared with cross correlation algorithm, the technique is expected to solve memory space problems by compressing all information from large cross-correlated pulse video into a single reconstructed image as an output. Furthermore, in existing cross correlation methods, the pulse peak time shifts with defect depth. Hence, defect quantification algorithms, such as SNR calculation, require multiple frame analysis. Such algorithms are comparatively simplified in sparse reconstruction technique. This paper explores sparse reconstruction algorithm for resolving close-spaced defects. This paper further describes cross-validation method for optimization of a user parameter in sparse reconstruction method.