可重构星球探测机器人的运动学建模及轨迹规划

提出了可重构星球探测机器人的概念,对系统中子机器人的研究进行了重点论述.通过设计恰当的子机器人连杆坐标系,利用Denavit Hartenberg方法完成了子机器人的运动学建模,并直接给出了子机器人的运动学正解模型.由于使用单一的求解算法不能求出工作空间的封闭解,因此综合利用代数法、几何法原理及空间投影关系,结合子机器人的结构特殊性推导出了运动学逆解,从而得到了工作空间内的所有解.在此基础上,考虑结构间的约束关系,给出了子机器人的工作空间及轨迹规划方法.最后,使用OpenGL对设计的子机器人系统进行了运动学仿真实验,实验以末端操作器的直线运动为例,充分考虑空间几何的关系,其结果有效地证明了建模及轨迹规划的正确性.

统计不相关最佳鉴别矢量集的本质研究

对统计不相关最佳鉴别矢量集的本质进行研究 ,在基于总体散布矩阵特征分解的基础上 ,构造了一种白化变换 ,使得变换后的样本空间中的总体散布矩阵为单位矩阵 ,这样使得传统的最佳鉴别矢量集算法得到的均是具有统计不相关的最佳鉴别矢量集 ,从而揭示了统计不相关最佳鉴别变换的本质———白化变换加普通的线性鉴别变换。该方法的最大优点在于所获得的最优鉴别矢量同时具有正交性和统计不相关性。该方法对代数特征抽取具有普遍适用性。用ORL人脸数据库的数值实验 ,验证了该方法的有效性

一种新的图像特征抽取方法研究

对最佳鉴别矢量的求解方法进行了研究，根据矩阵的分块理论和优化理论，在一定的条件下，从理论上得到类间散布矩阵和总体散布矩阵的一种简洁表示方法，提出了求解最佳鉴别矢量的一种新算法，该算法的优点是计算量明显减少。ORL人脸数据库的数值实验，验证了上述论断的正确性。实验结果表明，虽然识别率与分块维数之间存在非线性关系，但可以通过选择适当的分块维数来获得较高的识别率。类间散布矩阵和总体散布矩阵的一种简洁表示方法适合于一切使用Fisher鉴别准则的模式识别问题。

复杂条件地层滤波预测的理论与关键技术研究

The theory researches of prediction about stratigraphic filtering in complex condition are carried out, and three key techniques are put forward in this dissertation. Theoretical aspects: The prediction equations for both slant incidence in horizontally layered medium and that in laterally variant velocity medium are expressed appropriately. Solving the equations, the linear prediction operator of overlaid layers, then corresponding reflection/transmission operators, can be obtained. The properties of linear prediction operator are elucidated followed by putting forward the event model for generalized Goupillaud layers. Key technique 1: Spectral factorization is introduced to solve the prediction equations in complex condition and numerical results are illustrated. Key technique 2: So-called large-step wavefield extrapolation of one-way wave under laterally variant velocity circumstance is studied. Based on Lie algebraic integral and structure preserving algorithm, large-step wavefield depth extrapolation scheme is set forth. In this method, the complex phase of wavefield extrapolation operator’s symbol is expressed as a linear combination of wavenumbers with the coefficients of this linear combination in the form of the integral of interval velocity and its derivatives over depth. The exponential transform of the complex phase is implemented through phase shifting, BCH splitting and orthogonal polynomial expansion. The results of numerical test show that large-step scheme takes on a great number of advantages as low accumulating error, cheapness, well adaptability to laterally variant velocity, small dispersive, etc. Key technique 3: Utilizing large-step wavefield extrapolation scheme and based on the idea of local harmonic decomposition, the technique generating angle gathers for 2D case is generalized to 3D case so as to solve the problems generating and storing 3D prestack angle gathers. Shot domain parallel scheme is adopted by which main duty for servant-nodes is to compute trigonometric expansion coefficients, while that for host-node is to reclaim them with which object-oriented angle gathers yield. In theoretical research, many efforts have been made in probing into the traits of uncertainties within macro-dynamic procedures.