Two-dimensional ultrasonic flaw detection based on the wavelet packet transform

An important issue in ultrasonic nondestructive evaluation is the detection of flaw echoes in the presence of coherent background noise associated with the microstructure of materials. Many signal processing techniques have proven to be useful for this purpose, but fully 2-D flaw detection techniques remain desirable. In this paper, we describe a novel automatic flaw detection method based on the wavelet packet transform, which is particularly well adapted to B-scan image analysis. After a brief review of the essential elements of the theory of wavelets and wavelet packets, a detailed description of the method is provided. The detection process operates on a set of spatially oriented frequency channels, i.e., detail images, obtained from successive wavelet packet decompositions of the initial B-scan. A statistical selection procedure based on the modeling of the detail image histograms retains the useful information-bearing frequency channels. The flaw information is then extracted from these selected channels by means of a specific thresholding scheme. Some experimental detection results in B-scan images of austenitic stainless steel samples comprising artificial flaws are presented.