【作者】 王浩全；

【导师】 韩焱；

【作者基本信息】 中北大学 ， 精密仪器及机械， 2011， 博士

【摘要】 超声波具有无电离辐射、对人体无害、发射/接收设备价格便宜等特点,在无损检测领域得到了广泛应用。超声层析成像检测由于能够通过彩色或灰度图像直观地反映被测材料内部的质量,对缺陷进行定性定量分析,已逐渐成为研究的热点之一。针对超声层析成像检测中,获取的数据不完全、干扰因素多、误差大等缺点,本文提出了一种基于环绕式的阵列成像检测方法。该方法具有两大优点:(1)可以布置成任意形状。既可以根据试件的形状来布置,也可以根据重建算法的难易程度来布置;(2)超声波所经过的区域更加广泛,获得的数据更加全面。在材料内部衍射较弱、只考虑折射现象的情况下,将射线理论引入到超声层析算法中。针对线性插值射线追踪算法中发射点、接收点和缺陷都处于同一列(行)时得到的路径难以绕开缺陷区的缺点,提出了一种基于交叉扫描的射线追踪算法。该算法采用交叉扫描的方式进行前向处理,引用离散点作为次级元对后向处理过程进行简化。通过对路径的追踪证明新的算法能够有效地解决线性插值射线追踪算法中存在的问题。在获得超声最小走时和最短追踪路径的基础上,利用最小二乘方准则,对SIRT重建算法进行了推导。在相关系数矩阵的求取方面,提出了四边扫描算法。该方法利用像素的四条边所在直线与实际超声波的交点,通过判断交点与像素四个边界点的大小关系来实现系数矩阵的求取,使得系数矩阵求取更加方便、快捷。通过实验研究表明:采用环绕式阵列成像方法,在求取系数矩阵的基础上,利用联合迭代重建算法,在同样的迭代次数下,获得的数据更加精确,成像效果更好。在材料内部衍射较强的情况下,从超声波在连续介质中的传播特性出发,建立了关于精确散射场和全场的波动方程,运用矩量法实现了散射场和全场方程的离散化。基于环绕式阵列成像方法对实验模型进行了设计,利用Picard准则对实验模型能否进行正则化进行了判别。运用变形的Born迭代解决了波动方程的非线性问题,利用截断奇异值分解正则化方法解决了波动方程的稳定性问题。经实验证明本文所采用的基于空间域层析成像方法可以运用到较强散射的情况,可以反演较高的对比度,而且具有更高的成像质量。在讨论小波阈值降噪方法的基础上,分析了Dohono软、硬阈值化方法存在的缺点,设计了一种基于双曲线的新小波阈值函数。该阈值函数通过对参量的调节,可以获得较优的小波系数的阈值估计,能够克服硬阈值函数不连续和软阈值函数有偏差的缺点。针对超声层析图像的特点,基于区域分割原理,利用灰度直方图区域生长提供生长准则,同时改变分割方向,得到了一种新的分割思路。基于区域生长与模糊聚类分析方法,提出了区域生长与聚类分析相结合的分割算法,该算法在处理图像中由于信息不足而产生的模糊不确定问题时更加符合实际。更多还原

【Abstract】 Ultrasound has such characteristics as no ionizing radiation, hurtless in available frequency scale and inexpensive transmitter-receiver equipment, so it has great application potential in nondestructive testing. The ultrasond computed tomography testing is gradually becoming a hotspot in research because it can reflect the inner quality of materials by color or grayscale images.Creating a tomographic device with ultrasound is a new target for researchers.The linear array has some shortcomes such as lack of data,too much interference factor, big error and so on. A method of surround array is proposed according to the shortcomes.The method has two merits:(1) Sensors can be arranged with random shape;(2) The area which ultrasound goes through is more extensive.The acquired data is more .Ignoring the weak diffraction in materials, we introduced the ray theory into the algorithm of ultrasound computed tomography.The linear travel-time interpolation ray tracing algorithm has a drawback: rays can’t go round the low velocity area if the launching point, the receiving point and the low velocity area distribute on the same column (row). Aiming at the drawback, an improved algorithm is proposed on the basis of researching reconstruction theories. The given algorithm adopts a cross-scanning strategy to carry out forward processing and uses discrete points as the secondary sources to simplify backward processing. The numerical simulation experiments show that the rays can go round the low velocity areas located in the same column (row) of the launching points and the receiving points. The simultaneous iterative reconstruction techniques is derived through the least squared criterion. The algorithm of four boundaries scanning is proposed to get correlation matrixes.We can acquire the correlation matrix faster and more convenient by judging the relationship between the intersections and the four boundary points of the pixel.The experimental results show that the images can be more precise through SIRT algorithm in the same iterations.In the situation of big relative density about diffraction, a precise wave equation is deduced based on the propagation characteristics of ultrasound. The discretization algorithm about the scattering filed and the whole field is discussed with the method of momen.The expersimental model is designed based on surround array.Whether the model can be regularized is judged with the Picard criterion. The nonlinear of equations is solved with distorted Born iIterative.The ill-posedness of equations is solved with TSVD regularization. As a result, the convergence rate is accelerated. The imaging in spatial domain is no longer based on a approximation.The method can be applied in strong scattering field. The images are clearer. On the basis of wavelet threshold noise reduction, we analyzed the shortcomings about soft and hard thresholding of Dohono.A new wavelet threshold function based on hyperbola is proposed. It can acquire better estimate of wavelet coefficient, overcoming the weakness of uncontinuous hard threshold function and imprecise soft threshold function. A new idea of image segmentation is adopted based on image histogram and region growing.A new algorithm of image segmentation based on fuzzy mathematics and clustering analysis is applied to ultrasound CT images. It better solves uncertain question in the process of image processing.更多还原