Study On Terahertz Spectrum Recognition Based On Deep Belief Network

Terahertz technology is an interdisciplinary which develop rapidly in recent years. Because terahertz has many unique advantages, it has caused the extensive concern and research. At present, part of terahertz products have been used in industrial production, which has a great application potential and relatively high popularizing use-value. Because terahertz spectroscopy has the "fingerprinting" feature, the identification of terahertz spectroscopy has become an important way to identify substances, particularly for a macromolecules (such as drugs, herbs, crops, food, explosives, etc.). However, with the rapid development of terahertz time-domain spectroscopy technology and the improvement of equipment detection accuracy, which lead to the number of samples terahertz spectroscopy substances sharp growth trend, how to effectively use these data is a major problem which faced in the terahertz spectroscopy identification field.Compared to define and select feature artificially, Deep Belief Network (DBN) is a method of extraction feature automatically, which re-emergence from the beginning of 2006 and effectively deal with large-scale data. In the fields of natural language processing, voice and image have achieved great success. Because many materials have no apparent absorption peaks in the terahertz band, it is difficult to extract theirs terahertz spectroscopy feature and identify. To this end, a novel of identify terahertz spectroscopy approach with DBN was studied in this paper, which combines the advantages of DBN and K-Nearest Neighbors (KNN) classifier. Firstly, cubic spline interpolation and S-G filter were used to normalize the different kinds of substances terahertz transmission spectra. Secondly, the DBN model was built by two restricted Boltzmann machine (RBM) and then trained layer by layer using unsupervised approach. Instead of using handmade features, the DBN was employed to learn suitable features automatically with raw input data. Finally, a KNN classifier was applied to identify the terahertz spectrum. This proposed method shows a promising future in the application of identifying the mass terahertz spectroscopy.For almost all of the terahertz spectral database can only provide basic text information retrieval that use keywords to retrieve the corresponding spectrum, which lack spectrum retrieval function. Therefore, we explore a method of construct terahertz spectroscopy database based on Locality-sensitive Hashing (LSH) in this paper, the spectrum detection method were also studied as well. In order to validate the effectiveness of the method, a prototype terahertz spectroscopy identification system was built. Experimental results show that using this approach can reduce the search time greatly.