摘要

A novel method of time-frequency analysis, wavelet-packet atomic decomposition based on the matching-pursuit (MP) algorithm, is proposed for improvement of ultrasonic flaw detection during ultrasonic NDT. The MP algorithm is used to decompose a nonlinear and nonstationary signal into given atoms in an overcomplete wavelet-packet dictionary. The wavelet-packet dictionary with Daubechies wavelet functions, which well match the observed ultrasonic flaw echoes, is selected. By means of adaptively selecting the optimum atoms to reconstruct a signal, it is possible to obtain spare approximations of the original signal with less complexity and to efficiently improve a signal corrupted by noise. A distinct SNR enhancement for an ultrasonic echo during flaw detection is verified by computer simulation and experimental results.