| 何继爱,王倩宇,王志文.基于Dice 匹配的改进型匹配追踪算法[J].高技术通讯(中文),2023,33(5):459~466 |
| 基于Dice 匹配的改进型匹配追踪算法 |
| An improved matching pursuit algorithm based on Dice matching |
| |
| DOI：10. 3772/ j. issn. 1002-0470. 2023. 05. 002 |
| 中文关键词: 压缩感知(CS); Dice 系数匹配; 二次筛选; 选择性回溯 |
| 英文关键词: compressed sensing (CS), Dice coefficient matching, secondary screening, selective backtracking |
| 基金项目: |
| 作者 | 单位 | | 何继爱 | (兰州理工大学计算机与通信学院 兰州730050) | | 王倩宇 | (兰州理工大学计算机与通信学院 兰州730050) | | 王志文 | (兰州理工大学计算机与通信学院 兰州730050) |
|
| 摘要点击次数: 1060 |
| 全文下载次数: 811 |
| 中文摘要: |
| 内积匹配准则作为一种搜索最匹配原子的方法,被广泛应用在传统压缩感知(CS)算法中。然而,由于该准则无法对相似向量进行准确度量,通常会导致最匹配原子的误判率高,无法满足更高精度的数据重构需求。针对这一问题,本文提出一种基于骰子(Dice)匹配的二次筛选选择性回溯匹配追踪( DSS-SBMP) 算法,引入Dice 系数匹配准则解决内积匹配准则对两向量间相似度度量不准确的问题;通过对原子进行二次筛选来减少原子所对应支撑集内的错误索引数,同时引入选择性回溯克服迭代过程中存在的回溯过度现象。仿真结果表明,DSS-SBMP 算法在迭代过程中能够保留更多的正确原子,算法迭代次数小于子空间追踪(SP)算法,重构性能优于同类贪婪算法。 |
| 英文摘要: |
| The inner product matching criterion, as a method of searching for the best matching atom, is widely used in the traditional compressed sensing (CS) algorithm. However, owing to the low precision measure of similarity between vectors, the inner product matching criterion generally leads to high misjudgment rate of the best-matched atoms and can not meet the requirement of higher precision data reconstruction. To solve the above problem, secondary screening-selective backtracking matching pursuit based on Dice matching (DSS-SBMP) algorithm is proposed,which introduces the matching criterion of Dice coefficients to solve the problem that the matching criterion of inner product is inaccurate in measuring the similarity between vectors, the number of false indexes in the corresponding support set is reduced by the secondary screening of atoms, meanwhile, selective backtracking is introduced to overcome the backtracking excess in the iterative process. Simulation results show that DSS-SBMP algorithm can retain more correct atoms in the iterative process, the number of iterations is less than subspace pursuit (SP) algorithm,and the reconstruction performance is better than similar greedy algorithms. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
