Basic research on elimination of artifacts in ultrasonic imaging based on spatiotemporal CMRF model

超声成像在医学诊断、潜艇测量等各个应用领域具有广泛的优势。然而，超声波在传播过程中往往会受到多径效应、折射、绕射和旁瓣等伪影的影响。本研究的目的是通过复值神经网络(CVNN)对观测到的超声图像进行处理，为自适应识别和消除伪影奠定基础。其基本思想包括复值马尔可夫随机场(CMRF)模型和我们提出并多年来一直研究的CVNN。CVNN是神经网络的一种扩展，使我们能够自适应地、灵活地用显式的相位变量来处理与波有关的现象。为了识别和消除伪影，我们采用CVNN对声学成像中获得的复值图像数据进行处理。本项目取得了以下显著成果：(1)系统…更多的结构，我们在其中观察反射和/或伪连续波的散射的复幅度：我们已经构建了一个超声波表面测量系统，利用该系统，我们假设声阻抗突变的地方有一个清晰的边界，这是海底地貌观测的常见情况。发射50波包，对接收到的波形进行A/D转换和傅立叶变换，精确地得到载波频率分量的幅值和相位。(2)提出并论证了一种新的消除相位奇异点的方法：从上述系统获得的数据构造相位图时，发现其中包含密集的相位奇异点(SP)。我们需要减少SPS并展开阶段。首先，我们提出了一种新的相位解缠方法，即“扩展奇点位相解缠”。与常规方法相比，该方法具有计算成本低、质量高等优点。其次，我们提出了一种新的SP还原方法，即“基于CMRF的PS恢复方法”。我们在各自的SP上应用了该方法，以使它们移动到对应的SP以被抵消。结果表明，该方法实现了较大的缩减率和较高的相位预留，直接导致了高质量的DEM生成。较少

Ultrasonic imaging provides us with a wide range of advantages in various application fields such as medical diagnosis and submarine measurement. The ultrasonic wave is, however, often affected by the artifacts caused by multipath effect, refraction, diffraction, and sidelobes in the wave propagation. The aim of this study is to establish a basis to identify and eliminate the artifacts adaptively by processing observed ultrasonic images with complex-valued neural networks (CVNNs). The basic idea includes the complex-valued Markov random field (CMRF) model and the CVNNs, which we proposed and have been investigated for these years. The CVNNs is an extension of neural networks so that we can deal with wave-related phenomena adaptively and flexibly with explicit phase variables. We adopt the CVNNs to treat complex-valued image data obtained in acoustic imaging for the identification and elimination of the artifact. This project have yielded the following remarkable achievements.(1)System … More construction in which we observe complex-amplitude of reflection and/or scattering of pseudo-continuous wave: We have constructed an ultrasonic surface measurement system with which we assume a clear boundary where the acoustic impedance changes abruptly, which is an ordinary case of seabed landscape observation. We transmit 50-wave packet and detect the received waveform, which is A/D converted and Fourier transformed, and we obtain the amplitude and phase of the carrier frequency component precisely.(2)Development and demonstration of a novel method of phase singular point elimination: When we construct a phase map from the data obtained in the above system, we find dense phase singular points (SPs) included there. We need to reduce the SPs and unwrap the phase. First we proposed a new method to unwrap the phase, namely, "spread singularity phase unwrapping." The calculation cost is lower and the quality of the digital elevation map (DEM) is higher than those of conventional methods. Secondly we developed a novel SP reduction method, i.e., "CMRF-based PS restoration method." We applied the method at respective SPs to make them move to their counterparts to get canceled out. It has been found that the method realizes both a large reduction rate and an accurate phase reservation, which lead directly to high quality generation of DEM. Less