A Near Infrared Spectroscopy Model Transfer Method Based on Wavelet Transform Combined with Dynamic Time Warping

KeyWord:near infrared spectroscopy(NIR)  model transfer  wavelet transform  dynamic time warping

Abstract:

      In order to solve the problem of poor generality of near infrared spectroscopy(NIR) model caused by the difference of measuring environment and spectroscopy instruments,a model transfer method based on wavelet transform combined with dynamic time warping(WDTW) algorithm was proposed to realize the sharing of models among different instruments.First,the infrared spectra were preprocessed with wavelet transform.Then,the best match wavelength point relationship between two spectroscopy was acquired by using WDTW algorithm to build the regression model,and the performance of the developed model transfer method was verified with the experiment built transfer model using the near infrared spectroscopy dataset from drug and gasoline samples.The spectral standard deviation of prediction(SEP) for the gasolin datasets C7,C8,C9 and C10 were 0.414 4,0.801 1,1.090 4 and 1.290 8,respectively.The spectral SEP in activity,hardness and weight components of the drug datasets were 2.585 6,0.434 5 and 2.270 3,respectively,which were lower than those of the traditional methods.The experimental results showed that the WDTW model transfer algorithm has low standard error of prediction.It could not only eliminate the difference between the master and slave machine spectroscopy,but also improve the stability and accuracy of the model.