项目名称: 子空间学习粒子群算法及在图像过完备稀疏分解上的应用
项目编号: No.61261039
项目类型: 地区科学基金项目
立项/批准年度: 2013
项目学科: 无线电电子学、电信技术
项目作者: 孙辉
作者单位: 南昌工程学院
项目金额: 40万元
中文摘要: 本项目研究子空间学习粒子群优化算法模型,以解决算法在高维复杂问题中的早熟现象、提高算法的效率,并将其应用到图像过完备稀疏分解中,设计出快速、高效的图像过完备稀疏分解算法。涉及的研究内容包括:(1) 探讨度量粒子子空间进化优劣性质的评价指标,研究学习样本库的构建和管理规则,提出基于样本库的子空间学习策略和学习效果评价方案。(2) 研究子空间学习、参数调控和子空间大小调整等策略的自适应,使算法能够动态地调节,减少算法对问题的依赖,提高算法的适应性。(3) 借鉴树结构自适应非参数回归的思想,建立基于自适应空间分解的过完备稀疏分解模型,借助子空间学习粒子群优化算法,研究基于字典子空间学习的粒子群优化最匹配原子快速搜索算法,实现图像的快速过完备稀疏分解。 本项目的研究丰富了粒子群优化算法理论,具有重要的理论意义;同时为过完备稀疏分解提供了新的方法,具有重要的应用价值。
中文关键词: 粒子群优化算法;子空间学习;自适应;过完备;稀疏分解
英文摘要: The project studies particle swarm optimization model based on subspace learning to solve the premature convergence of the algorithm in high dimensional and complex problems, improve the efficiency of the algorithm, and applies it to the image overcomplete sparse decomposition to design a fast, efficient image overcomplete sparse decomposition algorithm. The involved research includes: (1) To explore evaluation indicators to measure the pros and cons nature of particle subspace evolution, to research the building and management rules of a learning sample database, and to propose subspace learning strategies based on sample database and learning effects evaluation indicators. (2) To study adaptive strategies of subspace learning, parameters regulating and subspace size adjusting, to enable the algorithm to dynamically adjust, reduce the dependence of the algorithm on the problem and improve the adaptability of the algorithm. (3) Drawing on the thinking of the tree structure adaptive non-parametric regression, to establish the overcomplete sparse decomposition model based on adaptive space decomposition. With the particle swarm optimization based on subspace learning, to study particle swarm optimization best match atom fast search algorithm based on dictionary subspace learning, to achieve image fast overcomplete
英文关键词: particle swarm optimization;subspace learning;self-adaptive;overcomplete;spare representation