Terahertz Spectrum Retrieval Based on P-stable LSH

LI Lingjie; SHEN Tao; NI Jiapeng

doi:10.3969/j.issn.1672-9722.2017.06.006

Abstract：The nearest neighbor search method of terahertz spectrum is a key problem in the full use of terahertz spectroscopy and related research. At present,there is no THz spectrum database which can provide spectrum search by THz spectrum. There?fore,in this paper,combined with the advantages of hash algorithm based on random projection,a new method of terahertz spec?trum retrieval based on P-stable distribution local sensitive hash algorithm is proposed. First of all,through the S-G filter and 3 spline interpolation and resampling,16 kinds of substances were normalized in the terahertz transmission spectrum of 0.9THz~6THz,and then the random projection method P-stable LSH is used to train the above original terahertz spectrum data in high di?mension spectrum based on the dimensionality reduction mapping to Hamming space,into the size of only tens of bits hash encod?ing,finally,parts of the spectrum of each substance selected as a query to retrieve and calculate the Hamming distance of hash code between query point and spectral data database,then the results are sorted and returned. Experimental results show that the retriev?al results by locality sensitive hashing algorithm using P-stable is more accurate and more efficient,the mean average precision (MAP)value is higher than other representative hash algorithms based on random projection.