Research On Terahertz Spectrum Recognition Based On Dictionary Learning

Electromagnetic wave in the Terahertz band does not exist the ionization properties of X ray,so it will not danage the material properties,and the photon energy is very low,so it can be used for nondestructive testing.As we all know,terahertz spectrum signals contains a large number of chemical and physical information,can provide the only fingerprint spectrum of substances with“fingerprint spectrum,characteristics,therefore can be used for material identification.So how can we make full use of the information contained in the signals for material identification effectively,how to use the terahertz spectrum data for material identification effectively,has become a big problem in the current field.With the emergence of sparse representation,it not only provides new ideas for signal representation,but also points out the new way of thinking for the exploration and research of terahertz time-domain spectroscopy signal.At present,the research on the terahertz signal representation is still in its primary stage,the traditional way of signal representation is not always effectively in reflecting the essential feature of the terahertz signal.To address the problems above,this paper presents a sparse representation method of terahertz signals,and due to the difficulty of terahertz spectrum feature extraction and classification,then proposes the method of terahertz signal recognition based on sparse representation.We had known that there are many questions in the traditional dictionary learning model.That is,the correlation of the analysis dictionary P is larger between classes,the sample code coefficients in the same class have great dispersion.Therefore,in this paper,we propose an improved dictionary learning model which is called the Regular Diagonal Structure of analysis Dictionary(DPL_RDS),by introducing a standardized diagonal matrix,not only reduces the relevance of the analytic dictionaries for each category but also reduces the dispersion of intra-coding coefficients.The experimental results show that the improved DPL_RDS algorithm has better performance than other algorithms,and the recognition effect is more obvious.Achieved the purpose of identification of substances with the terahertz spectrum.And the detection and identification of chemical substances are met by the use of the spectrum of the material.