Study On Data Processing Algorithm For Broadband Micro Spectrometer

A spectrometer is an instrument which measures and studies the intensity,wavelength,and associated variation of light radiation.Because of its high analytical accuracy,it is widely used in medical,metallurgical,environmental protection,astronomy and other fields.However,the big volume,high cost and inconvenience to carry of the traditional spectrometer limits its further popularization.Therefore,miniaturization becomes the inevitable trend of its development.In addition,due to the limitations of the spectral range and resolution,there is a problem that spectral resolution is not uniform and not high of the broadband micro spectrometer.Another disadvantage is that its ultraviolet band sensitivity is not ideal.Fortunately,the spectral data processing algorithm can improve the above two problems to a certain extent,and can improve the accuracy of the output spectrum of the spectrometer.In this paper,a wide-band micro-spectrometer is designed.Besides,the spectral data processing algorithm is further studied.Mainly for the following works:(1)The broadband micro spectrometer system is designed as a whole,including optical systems,hardware systems and software systems.Select the CCD as a photodetector,blazed grating as a spectroscope,and design the symmetrical Czerny-Turner structure as the optical system.Focusing on the design of optical components,the selection of optical components which including grating,collimator,imaging mirror,takes consideration of the size limit,aberration size,measurement band and resolution and other factors.(2)In order to eliminate optical and circuit noises in the spectrometer measurement process,the spectral signal denoising algorithm is studied.Firstly,the wavelet threshold de-noising theory is used to experiment.According to the Signal to Noise Ratio,Root Mean Square Error and Normalization Cross Correlation,it is found that the db wavelet group,Soft Shrinkage and Sqtwolog Threshold are suitable for spectral analysis.Secondly,to resolve the problem of denoising untotally or signal distortion in the process of threshold denoising,the singular value decomposition algorithm is introduced,which is combined with the wavelet algorithm.Finally,the experimental verification of the improved algorithm is carried out.The wavelet shrinkage is used to denoise the low-level wavelet coefficients of the high-frequency signal and the singular value decomposition to denoise its whole wavelet coefficients.The singular value denoising order is selected by the difference method.The experimental results show that denoising effect is complete and can retain the spectrum of effective features.It is proved that wavelet threshold-singular value difference method has a certain feasibility for denoising process of broadband micro spectrometer.(3)The calibration algorithm is studied from two aspects:wavelength calibration and intensity calibration.First,as the basis of wavelength calibration,the spectral peak finding algorithm uses derivative method.Secondly,based on the least squares theory,the mercury lamp wavelength calibration algorithm is studied,and the effects of first order,second order and third order polynomials are compared.The experimental results show that the absolute value of each fitting wavelength errors using the third order polynomial is less than 0.16nm.The residual square sum is 0.049638 and the standard error is 0.111398,which can meet the requirement of accuracy.The correlation index is almost close to 1.The degree of fitting is good,and is obviously better than the first and second order polynomial fitting.Moreover,within the whole spectral line(200-1100nm),the resolution of the spectrum after the mercury lamp calibration is uniform,all within 0.5nm,indicating that the broadband micro spectrometer has good spectral characteristics.Finally,the intensity calibration algorithm based on cubic B-Spline principle is studied.And the experiment of deuterium lamp intensity calibration is carried out.The experimental results show that the fitting correlation index can reach 0.9997,and the fitting curve has good smoothness and local fitting effect.This algorithm is suitable for intensity calibration of the spectral discrete signal.