基于双稀疏域联合求解的高精度光谱恢复算法
投稿时间:2020-03-16  修订日期:2020-05-12  点此下载全文
引用本文:
摘要点击次数: 37
全文下载次数: 0
作者单位E-mail
刘世界 中国科学院上海技术物理研究所 liushjie163@163.com 
李春来 中国科学院上海技术物理研究所 lichunlai@mail.sitp.ac.cn 
徐睿 中国科学院上海技术物理研究所  
唐国良 中国科学院上海技术物理研究所  
吴兵 中国科学院上海技术物理研究所  
徐艳 中国科学院上海技术物理研究所  
王建宇 中国科学院上海技术物理研究所 jywang@mail.sitp.ac.cn 
基金项目:中国科学院青年创新促进会(2016218);国家自然科学基金项目(11941002);高分辨率对地观测系统重大专项(GFZX04014308)
中文摘要:基于压缩感知的光谱成像系统需要合适的算法解码采样数据才能得到最终的光谱成像数据,传统单稀疏域变换算法会带来光谱细节损失等问题,针对该问题,本文提出了利用双稀疏域联合求解的方法(JDSD),将信号分解为低频部分和高频部分,并针对不同频率信号特点分别进行稀疏恢复,进而解码求解以实现高精度恢复信号。在数据验证中,首先利用OMP算法在频域内对光谱信息轮廓进行恢复,利用IRLS算法在空间域内对光谱细节进行补偿,分析了不同稀疏变换对于参数设置的影响,测试了不同算法组合的JDSD对于测试数据的恢复结果。对于500种光谱数据测试表明,双稀疏域联合求解可将光谱恢复保真度大大提升,20%采样率情况下, SAM和GSAM指标由传统方法的0.625和0.515分别提升为0.817和0.659,80%采样率情况下,SAM和GSAM指标由传统方法的0.863和0.808分别提升为0.940和0.897。JDSD算法可以使得光谱吸收峰等细节特征得到高精度保持,对于基于光谱的特征分析、物质识别等应用具有十分重要的意义。
中文关键词:光谱成像  光谱特征恢复  计算成像  压缩感知
 
High-precision algorithm for restoration of spectral imaging based on joint solution of double sparse domains
Abstract:Compressed sensing-based spectral imaging systems need to decode the sampled data by a proper algorithm to obtain the final spectral imaging data. Traditional decoding algorithms based on single sparse domain transformation will lead to loss of spectral details. Addressing this problem, a solution using transformation of two sparse domains was proposed. A signal was decomposed into a low frequency part and a high frequency part, sparse restoration was performed according to the characteristics of different frequencies, and then decoding was performed to obtain high-precision restored signals. In data verification, the OMP algorithm was firstly used to restore the spectral information profile in the frequency domain, then the IRLS algorithm was used to compensate the spectral details in the spatial domain. The impact of different sparse transformations on parameter settings was analyzed, and the JDSD of different algorithm combinations was tested. Tests on 500 kinds of spectral data show that the joint solution of double sparse domains can greatly improve the fidelity of spectral restoration. With a sampling rate of 20%, the SAM and GSAM indexes are increased from 0.625 and 0.515 by traditional methods to 0.817 and 0.659, respectively. In the case of 80%sampling rate, the SAM and GSAM indexes are increased from 0.863 and 0.808 of traditional methods to 0.940 and 0.897, respectively. The JDSD algorithm can maintain high-precision details such as spectral absorption peaks, it is great significance for applications such as spectral-based feature analysis and substance identification.
keywords:Spectral imaging  spectral feature recovery  computational imaging  compressed sensing
  HTML  查看/发表评论  下载PDF阅读器

版权所有:《红外与毫米波学报》编辑部