MRI梯度预加重模块的分时复用设计

为获得更优的成像质量和更快的成像速度,磁共振成像（MRI）系统的梯度预加重模块需要具有更多的补偿通道和调节参数,常规预加重模块的设计方案使现场可编程门阵列（FPGA）面临巨大的资源消耗.为解决高性能梯度预加重模块的资源消耗大的问题,本文提出了一种基于分时复用技术的梯度预加重实现方案,以常规方案1/44的资源实现了11通道×4组参数的梯度预加重模块.将该模块用于0.35 T MRI系统,测试了补偿前后的涡流曲线和磁共振图像,结果表明该模块有效降低了系统的涡流,减小了磁共振图像中的涡流伪影.     

热传导界面问题的一种神经网络算法

本文讨论了一种单隐层神经网络算法在数值求解热传导界面问题中的应用。该算法设定含有神经网络函数的近似解满足初边值条件和Dirichlet 界面条件,通过求解由原方程导出的关于神经网络权重的离散优化问题来训练近似解中的神经网络,以使近似解逼近真解。文中也给出了一种基于随机梯度法思想的类随机梯度法来求解相应的离散优化问题。数值算例验证了算法的有效性。     

热镀锌钢卷力学性能GBDT预报模型

针对热镀锌钢卷力学性能预报建模条件属性选取难、预报精度不足的问题,研究了热镀锌钢卷力学性能梯度提升树(gradient boosting decision tree,GBDT)预报模型。利用互信息差算法综合评估工艺参数、化学成分和钢卷尺寸参数等条件属性的相对重要性以及属性之间冗余性,进行模型条件属性筛选;采用同分布原理进行样本划分,结合网格搜索法和交叉验证法优化模型参数,建立力学性能GBDT预报模型。将GBDT模型预报结果与随机森林(random forest,RF)、AdaBoost算法和BP神经网络的预报结果进行比较,比较表明GBDT模型优于其他模型,90%的数据样本预测的绝对误差小于14.24 MPa,94.6%的数据样本相对误差在6%范围内,具有更高的预测精度。     

Prandtl number effects in decaying homogeneous isotropic turbulence with a mean scalar gradient

Decaying homogeneous isotropic turbulence with an imposed mean scalar gradient is investigated numerically, thanks to a specific eddy-damped quasi-normal Markovian closure developed recently for passive scalar mixing in homogeneous anisotropic turbulence (BGC). The present modelling is compared successfully with recent direct numerical simulations and other models, for both very large and small Prandtl numbers. First, scalings for the cospectrum and scalar variance spectrum in the inertial range are recovered analytically and numerically. Then, at large Reynolds numbers, the decay and growth laws for the scalar variance and mixed velocity–scalar correlations, respectively, derived in BGC, are shown numerically to remain valid when the Prandtl number strongly departs from unity. Afterwards, the normalised correlation ρ_wθ is found to decrease in magnitude at a fixed Reynolds number when Pr either increases or decreases, in agreement with earlier predictions. Finally, the small scales return to isotropy of the scalar second-order moments is found to depend not only on the Reynolds number, but also on the Prandtl number.     

Single-layer broadband planar antenna using ultrathin high-efficiency focusing metasurfaces

Phase gradient metasurfaces(PGMS) offer a fascinating ability to control the amplitude and phase of the electromagnetic(EM) waves on a subwavelength scale, resulting in new applications of designing novel microwave devices with improved performances. In this paper, a reflective symmetrical element, consisting of orthogonally I-shaped structures, has been demonstrated with an approximately parallel phase response from 15 GHz to 22 GHz, which results in an interesting wideband property. For practical design, a planar antenna is implemented by a well-optimized focusing metasurface and excited by a self-designed Vivaldi antenna at the focus. Numerical and experimental results coincide well. The planar antenna has a series of merits such as a wide 3-d B gain bandwidth of 15–22 GHz, an average gain enhancement of 16 d B, a comparable aperture efficiency of better than 45% at 18 GHz, and also a simple fabrication process. The proposed reflective metasurface opens up a new avenue to design wideband microwave devices.     

Effect of gravity gradient in weak equivalence principle test

A high accuracy test of the weak equivalence principle(WEP) is of great scientific significance no matter whether its result is positive. We analyze the gravity gradient effect which is a main systematic error source in the test of WEP.The result shows that the uncompensated gravity gradient effect from the coupling term of the dominated gravity gradient multipole moment component q_(21) and the relative multipole field component Q_(21) contributes to an uncertainty of 1×10~(-11) on the E¨otv¨os parameter. We make a Q_(21) compensation to reduce the effect by about 20 times, and the limit of the test precision due to this coupling is improved to a level of a part in 10~(13).     

基于一维Otsu的多阈值医学图像分割算法

针对传统医学图像分割算法时间复杂度高、 分割精度低等问题, 提出一种基于一维Otsu的自动多阈值分割算法. 考虑到医学图像信息的复杂性, 引入基于梯度、 灰度、 距离的综合信息直方图替代传统的灰度直方图, 并分别赋予这3个信息相应的权值. 采用kd-树作为框架快速自动确定阈值个数, 进而实现Otsu对医学图像的自动多阈值分割. 与最大熵、 基于粒子群优化的Otsu算法等进行对比实验的结果表明, 该算法的分割性能优于其他算法.     

基于梯度法的高效全局优化算法

用极大似然估计法和交替方向法估计Kriging模型参数, 提出一种基于有效集共轭梯度法的Kriging模型参数优化算法, 并在此基础上改进了高效全局优化算法. 结果表明, 利用改进的全局优化算法可解决高效全局优化算法的过早收敛问题.     

基于共轭梯度迭代算法受控AR模型的参数辨识

推导了单输入单输出系统的辅助模型,它有助于减少计算量和提高共轭梯度迭代算法（新算法）的收敛速度.相比于受控移动平均模型中所提出的交互式随机梯度算法,新算法用更少的迭代步骤就可求出模型的参数估计.另外,新算法能避免出现矩阵的逆矩阵形式.对新算法与双共轭梯度算法进行比较,并给出数值实例检验新算法的有效性.     

PGA算法在条带式SAS场景目标成像中的应用

本文针对条带式合成孔径声呐(SAS)对场景目标成像应用中面临的方位向残余相位误差导致成像质量降低的问题,探讨了相位梯度自聚焦(PGA)算法在条带式SAS成像方位向相位误差估计和补偿中的应用方法,通过自适应的选点及窗宽估计处理以提高算法对场景目标的适应性,最后通过对场景目标成像的实际湖试数据进行处理和分析,结果验证了该算法改善条带式SAS对场景目标成像质量的有效性。