Performance Evaluation of 3D Compression for Ultrasonic Nondestructive Testing Applications Pramod Govindan and Jafar Saniie Department of Electrical and Computer Engineering Illinois Institute of Technology, Chicago, Illinois, U.S.A. Abstract- Ultrasonic imaging and nondestructive testing applications require large volumes of data. Therefore, it is essential to compress data for storage and transmission without degrading the signal quality or inadvertently damaging desirable signal features. In this study, the ultrasonic signal compression and signal reconstruction is implemented using discrete wavelet transform (DWT), and direct signal decimation and reconstruction by interpolation when sampling rate exceeds several folds above the Nyquist rate. The objective of this study is to design 1D and 3D compressions and evaluate the degree of compression of ultrasonic data as a function of data integrity. The compression performance is evaluated by calculating the PSNR (peak SNR) and the correlation between the original and reconstructed signals with respect to compression ratio. The performance of the 3D compression algorithm used in this study offers an overall 3D compression ratio of 95% with a PSNR of 27dB indicating high signal fidelity. I. INTRODUCTION One of the major challenges in ultrasonic signal processing application is the large volumes of data to be processed. The requirement of real-time processing of compute–intensive signal processing algorithms further increases the implementation challenges. Since the processed data may need to be transmitted to remote locations for expert analysis, diagnosis and archiving via low bandwidth communication channels, the data transmission time has to be reduced. Therefore, the ultrasonic signal has to be compressed as much as possible without degrading the signal reconstruction quality. For detecting small and transient features within the signal and getting a high time resolution, it is a general practice to sample the signals at much higher rate than Nyquist rate. This oversampling ensures a high signal-to-noise ratio (SNR). However, oversampling generates huge amount of data with redundant information. This will create storage problem and cause a large delay and/or bandwidth requirement in transmitting the data. Consequently, the signal needs to be compressed without sacrificing signal fidelity. One of the most common and classical methods for compression is Discrete Wavelet Transform (DWT), which is used for ultrasonic compression applications due to its high energy compaction properties [1]. By carefully selecting the most suitable wavelet kernel, maximum compression can be achieved using DWT. Another efficient method for compression is by decimation where the redundant information is removed, and this may cause degradation of time resolution. However, the decimated signal may need to be interpolated when certain signal features with fine time resolution are required. In this study, the shift-property of Fourier transform is used to recover the decimated samples and to reconstruct (i.e., interpolate) the signal. Furthermore, 3D compression of ultrasonic data is performed using DWT. The compression is applied to 3D block of data through successive 1D compression on each of the 3 directions (x, y and z). The similarity between neighboring A-scans indicates high data correlation, and this promotes the possibility of high compression ratio with high signal fidelity. This paper is organized as follows. Section II explains the DWT for 1D compression of ultrasonic signals. Section III provides an alternate compression method by using decimation and time-shift interpolation. Section IV describes the experimental setup for 3D ultrasonic data acquisition and the compression of the acquired 3D data using DWT. The performance analysis of 3D compression is also discussed in this section. Section V concludes this paper. II. DWT BASED COMPRESSION DWT is performed by transforming and decomposing the signal into lowpass and highpass coefficients. Maximum compression can be achieved if more coefficients can be identified with very little energy which can be eliminated. Wavelet packet decomposition [2] is used in our experiment, where apart from further decomposing the lowpass (L) components, selected highpass (H) components with high energy distribution are also decomposed. This approach improves the data compaction and leads to a higher compression ratio. The best compaction can be achieved by properly choosing the right wavelet kernel to decompose the ultrasonic signal. The kernel must be chosen such that most of the wavelet coefficients are small, which can be approximated to zero without affecting signal quality. This scenario depends on the regularity, number of vanishing moments of the scaling function of the wavelet kernel and the support size (i.e., time-width) of the kernel [3], [4]. 437 978-1-4673-5686-2/13/$31.00 ©2013 IEEE 2013 Joint UFFC, EFTF and PFM Symposium 10.1109/ULTSYM.2013.0113 Authorized licensed use limited to: Illinois Institute of Technology. Downloaded on October 02,2020 at 06:16:34 UTC from IEEE Xplore. Restrictions apply.