Compression for ultrasonic phased array imaging: compressive sensing and wavelet thresholding
R Pyle, R Bevan and P Wilcox
Advances in ultrasonic phased array imaging are making large improvements to defect detection and characterisation, but also massively increase the amount of data that must be stored. This is a huge challenge for applications such as structural health monitoring and in-line pipe inspection, where hours’ or even days’ worth of data must be stored. Traditionally in the NDE industry, properties such as time-of-flight and peak amplitude are stored and the signal discarded to overcome this issue. However, doing this removes the ability to use image-based detection techniques.
Lossy compression algorithms are a solution to this data size challenge as they can efficiently store signals while maintaining sufficient accuracy to image features of interest. This paper describes two such algorithms: compressive sensing (CS) and wavelet thresholding (WT). They are applied to simulated data of responses from point scatterers and their effect on reconstruction accuracy is compared using signal-to-noise ratio and normalised cross-correlation. Both methods succeed in reducing data size to at least four times smaller than the Nyquist criterion requires, while maintaining detection capabilities, but have relative merits. CS is shown to have lower computational complexity at the compression stage (where processing power is often limited), whereas in this case WT affects less error in the process.
Lossy compression algorithms are a solution to this data size challenge as they can efficiently store signals while maintaining sufficient accuracy to image features of interest. This paper describes two such algorithms: compressive sensing (CS) and wavelet thresholding (WT). They are applied to simulated data of responses from point scatterers and their effect on reconstruction accuracy is compared using signal-to-noise ratio and normalised cross-correlation. Both methods succeed in reducing data size to at least four times smaller than the Nyquist criterion requires, while maintaining detection capabilities, but have relative merits. CS is shown to have lower computational complexity at the compression stage (where processing power is often limited), whereas in this case WT affects less error in the process.
