计算机工程与设计
計算機工程與設計
계산궤공정여설계
COMPUTER ENGINEERING AND DESIGN
2015年
7期
1784-1789
,共6页
黄宏伟%谢正光%蒋小燕%蔡旭
黃宏偉%謝正光%蔣小燕%蔡旭
황굉위%사정광%장소연%채욱
压缩感知%原子%限制等距性质%残差%自适应算法
壓縮感知%原子%限製等距性質%殘差%自適應算法
압축감지%원자%한제등거성질%잔차%자괄응산법
compressed sensing%atom%restricted isometry property%residual%adaptive algorithm
为提高贪婪算法重构精度,提出带有回溯机制的基于限制等距性质阈值匹配追踪算法(restricted isometry proper‐tity‐based threshold mechanism MP ,RIPTMP)。每次迭代包含原子添加和原子删减两个步骤,在原子添加步骤中,根据RIP和残差能量条件添加原子;在原子删减步骤中,分析 RIP和残差条件,找出可能错误原子,原子选择过程是自适应的。实验结果表明,在一定条件下,该算法重构精度高于正交匹配追踪算法(orthogonal matching pursuit ,OMP)、子空间匹配算法(subspace pursuit ,SP)、基追踪算法(basis pursuit ,BP)和前向后向追踪算法(forward‐backward pursuit , FBP)等算法。
為提高貪婪算法重構精度,提齣帶有迴溯機製的基于限製等距性質閾值匹配追蹤算法(restricted isometry proper‐tity‐based threshold mechanism MP ,RIPTMP)。每次迭代包含原子添加和原子刪減兩箇步驟,在原子添加步驟中,根據RIP和殘差能量條件添加原子;在原子刪減步驟中,分析 RIP和殘差條件,找齣可能錯誤原子,原子選擇過程是自適應的。實驗結果錶明,在一定條件下,該算法重構精度高于正交匹配追蹤算法(orthogonal matching pursuit ,OMP)、子空間匹配算法(subspace pursuit ,SP)、基追蹤算法(basis pursuit ,BP)和前嚮後嚮追蹤算法(forward‐backward pursuit , FBP)等算法。
위제고탐람산법중구정도,제출대유회소궤제적기우한제등거성질역치필배추종산법(restricted isometry proper‐tity‐based threshold mechanism MP ,RIPTMP)。매차질대포함원자첨가화원자산감량개보취,재원자첨가보취중,근거RIP화잔차능량조건첨가원자;재원자산감보취중,분석 RIP화잔차조건,조출가능착오원자,원자선택과정시자괄응적。실험결과표명,재일정조건하,해산법중구정도고우정교필배추종산법(orthogonal matching pursuit ,OMP)、자공간필배산법(subspace pursuit ,SP)、기추종산법(basis pursuit ,BP)화전향후향추종산법(forward‐backward pursuit , FBP)등산법。
To improve the reconstruction accuracy of greedy algorithms ,an algorithm with backtracking mechanism called re‐stricted isometry propertity‐based threshold mechanism MP (RIPTMP) was presented .RIPTMP had two procedures in each iteration called atoms‐adding and atoms‐deletion .In the atoms‐adding procedure ,atoms were picked out based on the condition of RIP and the energy of signal residual .Meanwhile in the step of atoms‐deletion ,the property of RIP and signal residual was also studied to find those atoms that were possibly wrong .The whole procedure of atoms selection was adaptive .Experimental result shows that under certain circumstances ,the reconstruction accuracy of RIPTMP surpasses that of OMP ,SP ,BP ,FBP and some other greedy algorithm .