Open path Fourier transform infrared (OP–FTIR) spectroscopy is qualified in detecting mixtures by multivariate calibration methods such as partial least squares (PLS); however, its applications are still restricted by background noise, which is unavoidable for OP–FTIR spectra and cannot be resolved solely by multivariate calibration methods. Hence OP–FTIR spectra are often pretreated before the data are subjected to the multivariate calibration model. A new preprocessing technique, orthogonal signal correction (OSC), was presented in this paper. The principle of OSC is to remove the part in X orthogonal to Y, and it is implemented based on PLS and nonlinear iterative partial least squares (NIPALS) algorithm. The approach was applied to three different data sets of PLS model. It performed much better than classical PLS when handling data with noise but comparably when processing on the simulated data. Moreover, OSC could reduce the complexity of model, which would facilitate the interpretation of the models. The results reveal that the proposed method gives much better prediction than the classical PLS and is very promising for the wide use of OP–FTIR. The preprocessing technique, aut–oscaling, and second–order derivatives (SOD) were also considered.
Key Words :
ACKNOWLEDGMENTS
This project was supported by the National Natural Science Foundation of China (No. 20175008), by the Postdoctoral Foundation of Education Ministry of China, and by the Young Scholar Foundation of Nanjing University of Science and Technology (Njust200303).