| The accuracy of the micro spectrometer measurement is affected by hardware selection,optical design and environmental factors.Among them,bandwidth is an important factor affecting the precision of the micro spectrometer data,and bandwidth correction is one of the key technologies to improve the precision of the micro spectrometer.The influence of the bandwidth of the micro spectrometer on the measured spectrum was studied,and two intelligent bandwidth correction algorithms were designed to reduce the influence of the bandwidth on the measured spectrum.Firstly,the error and bandwidth correction principle of the micro spectrometer were analyzed from the physical aspect,and several classical bandwidth correction methods were introduced.The problems of least square method in bandwidth correction were analyzed and some regularization parameter selection methods were introduced.A bandwidth correction method based on adaptive Tikhonov(TV)regularization and Levenberg-Marquardt(L-M)algorithm was proposed,and the spectral parameters were optimized by this method.The comparison/verification experiments of light-emitting diode(LED)and Raman spectra under different TV regularization parameters demonstrate the superiority of L-M algorithm based on adaptive TV regularization.In order to generalize the parameter selection method to all the algorithms based on the least square method,a parameter selection method based on deep learning(DL)was proposed.A database and neural network were established,and the optimal parameters of the corresponding algorithm were obtained through the training of neural network.The distorted LED,Raman,and Compact Fluorescent LAMP(CFL)spectra were corrected using L-M and Richardson-Lucy(R-L)algorithms with optimal parameters,and compared/validated with traditional L-M and R-L algorithms.The A-type uncertainty and Root Mean Squared Error(RMSE)under different conditions were calculated.The experiment proves that the parameter selection algorithm based on DL can effectively select the optimal parameters,so that the corresponding algorithm can more effectively restore the distorted spectrum and improve the quality of spectral restoration. |