Analytical applications of Near-Infrared spectroscopy technique have grown dramatically in recent years. It is widely used in lots of areas, such as petrochemical industry, agriculture and food industry, textile industry, pharmaceutics industry and clinical medicine. The wide using of NIR should attribute to the data processing method based on chemometrics. The purpose of this paper focused on pretreatment, one of important steps in the data processing of NIR spectra.Orthogonal signal correction (OSC) is a new popular pretreatment method, and its basic idea is that it removes the parts linearly unrelated (orthogonal) to the response matrix Y from spectral matrix X. In this paper, OSC was applied to preprocess the NIR spectra of herbal medicines. The results show that OSC is suited to preprocess the NIR spectra of herbal medicines. Furthermore, six different algorithms of OSC were compared and the results show that OPLS is the most powerful method of them.The drawback of OSC was discussed in this paper. The noises in the spectra are not absolutely orthogonal to Y, so only the projection on the orthogonal direction to Y can be removed by OSC. The residuals of noises will influence the stability of models and may be a reason of over-fitting. An improvement method is proposed in this paper. Multiple chains method is used for wavelength selection before OSC, and the unrelated points to models are removed for eliminating the interference of noises in some extent. This method is applied to treat the NIR spectra of corn, and the accuracy of calibration model are improved. |