The neatest neighbor (NN) method is one of the most typical methods in spectral retrieval, automatic processing and data mining. The main problem in NN is the low efficiency. Therefore,focus on the efficient implementation problem and introduce a novel and efficient algorithm SHNN (sequential computation-based hash nearest neighbor algorithm). In algorithm SHNN, firstly, decompose and recognize the spectrum flux components based on their hashing power; Secondly, the nearest neighbor is computed in PC A space based on sequential computation idea. In the second procedure,the putative nearest spectra can be reduced based on hash idea,and the un-nearest spectra can be rejected as early as possible. The contributions of this work are; 1) anovel algorithm SHNN is introduced,which improve the efficiency of the most popular spectramining method nearest neighbor significantly;2) Its application in star spectrum,normal galaxy spectrum and Qso spectrum classificationis investigated. Evaluated the efficiency of the proposed algorithms experimentally on the SDSS (Sloan Digital Sky Survey) released spectra. The experimental results show that the proposed SHNN algorithm improves the efficiency of nearest neighbor method more than 96%. The nearest neighbor is one of the most popular and typical methods in spectra mining. Therefore , this work is useful in a wide scenario of automatic spectra analysis, for example, spectra classification, spectra parameter estimation, redshift estimation based on spectra,etc.%基于近邻的方法是海量光谱数据获取、自动处理和挖掘中的一类重要方法,在应用中它们的主要问题是效率较低,为此文中提出了基于序贯计算的散列近邻法( SHNN).在SHNN中,首先使用PCA方法对光谱数据进行正交变换,使数据按照各成分的散列能力进行组织；然后在PCA空间中快速查找待识别光谱的近邻数据,在此过程中通过散列思想快速约减搜索空间,并用序贯计算法高效地排除非近邻光谱数据,提高计算效率.文中主要贡献是,提出了SHNN算法,并研究了该算法在恒星光谱、正常星系光谱和类星体光谱识别中的应用.SDSS光谱实验研究表明,SHNN效率提高约96％以上,速度提高26.45倍以上.由于近邻法的广泛适用性,文中研究结果不仅对恒星光谱、正常星系光谱和类星体光谱的识别研究有重要的参考意义和一定的实用价值,亦对恒星大气参量的估计和基于光谱的红移测量有一定的参考意义.
展开▼
