项目名称: 一类非凸分裂可行问题及其应用研究
项目编号: No.11301253
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 数理科学和化学
项目作者: 王丰辉
作者单位: 洛阳师范学院
项目金额: 23万元
中文摘要: 分裂可行问题是一类重要的非线性问题,其在信号重建和图像恢复领域有着广泛的应用。目前非凸框架下分裂可行问题的研究尚处于起始阶段,本项目将以半代数集为切入点,深入探讨o-极小结构这类非凸集类上的分裂可行问题。首先,应用Clark广义微分理论,forward-backward分裂方法和次梯度投影方法的思想,构造出求解此类问题的基本算法框架。 其次,以渐进正则性,动力系统中的经典方法和Kurdyka-Lojasiewicz不等式为主要工具建立算法的收敛性和稳定性分析理论。最后,通过在信号重建和图像恢复领域中的数值实验检验算法的效率,分析影响算法收敛速度的各种因素。
中文关键词: 分裂可行问题;非扩张算子;迭代算法;强收敛性;变步长
英文摘要: The split feasibility problem is a class of nonlinear problems, which has wide application in signal processing and image recovery. However, the current research on such a problem under the non-convex setting is still in its primary stage. By using the semi-algebric set, we will study the problem under the o-minimal structure. Firstly, we will propose an iterative method based on Clark's subdiffential theory, the forward-backward splitting method and the subgradient projected algorithm. Secondly, we will establish the convergence and stability result of the proposed algorithms by using the asymptotic regularization, the classical methd in dynamical system and the Kurdyka-Lojasiewicz inequality. Finally, we will check their efficiency and analyze their effects on algorithm convergence through the numerical experiments in the siginal processing and image recovery.
英文关键词: split feasibility problem;nonexpansive operator;iterative algorithms;strong convergence;variable stepsize