A new method is proposed to eliminate the varying background and noise simultaneously for multivariate calibration of Fourier transform near infrared (FT-NIR) spectral signals. An ideal spectrum signal prototype was constructed based on the FT-NIR spectrum of fruit sugar content measurement. The performances of wavelet based threshold de-noising approaches via different combinations of wavelet base functions were compared. Three families of wavelet base function (Daubechies, Symlets and Coiflets) were applied to estimate the performance of those wavelet bases and threshold selection rules by a series of experiments. The experimental results show that the best de-noising performance is reached via the combinations of Daubechies 4 or Symlet 4 wavelet base function. Based on the optimization parameter, wavelet regression models for sugar content of pear were also developed and result in a smaller prediction error than a traditional Partial Least Squares Regression (PLSR) mode.

Date Published: 8 November 2005
PDF: 12 pages
Proc. SPIE 5996, Optical Sensors and Sensing Systems for Natural Resources and Food Safety and Quality, 599619 (8 November 2005); doi: 10.1117/12.630424

Published in SPIE Proceedings Vol. 5996:
Optical Sensors and Sensing Systems for Natural Resources and Food Safety and Quality
Yud-Ren Chen; George E. Meyer; Shu-I Tu, Editor(s)