Noise reduction in ultrasonic NDT using discrete wavelet transform processing

Discrete Wavelet Processing, with decomposition level-dependent threshold selection is used for the enhancement of the signal-to-noise ratio (SNR) of synthetic and experimental ultrasonic pulse-echo registers. An analysis of the SNR improvement is performed from the results obtained by processing synthetic grain noise registers having an incrusted flaw signal. Different threshold selection techniques, including Universal, Minimax and Sure rules, have been used. An alternative processing of the wavelet coefficients, with pruning and threshold selection based on the statistics of the squared coefficients at the particular scales is proposed.