Image reconstruction technique for ultrasonic transmission tomography

Abstract

Precise flow control has always been a necessity for developing easier approaches or instrumentation for two-phase flow regime. An important method for monitoring this process is called process tomography such as electrical tomography, optical tomography and ultrasonic tomography (UT). In the case of high-acoustic impedance mixtures e.g. bubbly flow, UT has the advantages in monitoring real time data. Although various researches were conducted using UT systems in bubbly flow regimes, there are still weaknesses especially in real time image reconstruction techniques for monitoring the process. Some efforts such as linear back projection (LBP), filter back projection (FBP), convolution back projection (CBP) and iterative techniques are utilized for reconstructing the image with few views data for UT system. Regardless of the utilized method there still exist two main issues in UT image reconstruction both in forward and inverse problems. In the case of forward problem, the gaps between sensitivity maps cause artifacts in a reconstructed image. Moreover, for inverse problem, limited number of sensors causes artifacts in reconstructed image. In the case of high noisy environment, the LBP, FBP and CBP methods are not capable of totally removing the noise and artifacts level. Dynamic motion of flow regime is considered as another issue in UT system which causes inaccuracy in image reconstruction. Therefore, these issues were considered in developing a modified image reconstruction algorithm which was based on improving the CBP algorithm both in forward and inverse problems. A modified sensitivity map based on Gaussian distribution was utilized to combat the gaps in forward problem, and for the case of inverse problem, the wavelet fusion technique was applied to reduce the noise level, artifacts and the effects of dynamic motions. The simulation and the experimental works had been conducted based on different static profiles. Various types of image reconstruction algorithms were implemented and compared with the proposed technique. The quality of the final reconstructed images was evaluated using structural similarity (SSIM) and peak signal to noise ratio (PSNR). Results show that the WCBP outperforms LBP and CBP in case of SSIM and PSNR. Comparing to LBP, the SSIM and PSNR were improved at least by 30% and 5% respectively while for CBP the improvement were about 5% and 1% respectively.