Researches Of Spectral Fitting And Baseline Correction Algorithm Based On Sparse Representation

Spectroscopy is a powerful method of material composition detection.The essence of this method is that a material absorbs part of the frequency of light when contact with light,leading to an energy-level leap of molecule or atom in the material.Spectroscopy has been widely used in agriculture,chemical industry,medicine,food,cultural relics testing and other fields as a non-destructive testing technology.It promotes the development of related industries greatly and facilitates human life greatly.There are usually noise interference and baseline drift interference in the process of obtaining material spectroscopic map,which affects the identification of spectral peek and height.It leads to the low accuracy of the composition of the subsequent spectral analysis,and it is difficult to apply in practice.Noise interference mainly comes from the process of converting analog signal stouor into digital signal and environmental factors,and baseline drift is mainly generated in the operation of the machines.The traditional methods only remove the noise or correct the baseline,which need to be performed step by step to complete the spectral pretreatment,resulting in poor final treatment effect.To address this shortcoming,it is based on the sparse representation of spectral signal to design a sparse model of spectral fitting and baseline correction at the same time in this thesis.Its main contents are as follows:1.Sparse constraint was applied to the spectral signal representation coefficient and of the signal.2.Add baseline smoothness constraints into the sparse model to make the solved3.It solves the baseline and spectral representation coefficients by separable and removes the noise implicitly.4.Not satisfied with the convergence speed of detachable alternative functions,the thesis to improve the convergence speed.Because of the separation ofl₁ norm andl₂ norm in the designed model of this thesis,the algorithm can handle multiple spectral signals at the same time and gets great advantages for large-scale spectral signal processing.The experimental results show that the algorithm has achieved good results on the simulated spectral data set and on the real spectral data set.