The Signal Features Extraction Of Near-infrared Spectroscopy

There is a series of puzzles in Near-Infrared(NIR) Spectroscopy of Inspector Technology ,the core of which is that the near-Infrared spectrum, as the sources of information, has low effective information and to analyze the complex samples with NIRS is to extract slender information from complicated, overlapping, alternant spectrum. So extracting effective information from NIR spectrum and optimizing the original spectrum data are crucial to develop ideal mathematical model in NIRS. From both macroscopic and microscopic views, this paper, taking Metronidazole as the object of study, uses signals analysis method and data analysis theory to extract characteristic information from NIR spectrum. The main contents and methods of this paper are:1. Collecting spectrum of different content samples with high distinguish-ability, the original spectrum with abundant information are obtained.2. With multi-scale edge extraction theory and Wavelet transform method, multi-scale analysis and local magnitude maxima description of spectrum is set up.3. Using correlation analysis technology to compare the relation curve of mid-IR spectrum obtained from different instruments and different operating environments .So to study the practicability of this method.4. Through cross correlation analysis between pure Metronidazole and different content samples, analyze the actual effect of this method. Based on the correlation of pure components, extracting characteristic information of the multi-component samples.5. Applying Partial Least-Squares method to characteristic data obtained from the two methods above, setting up linear mathematical model of correlation between NIR spectrum and Metronidazole content. Meanwhile, using predicted data to make prediction effect analysis. So to verify the two methods.The following conclusions are obtained according to the research described above: both multi-scale edge extraction based on Wavelet transform and correlation analysis technology are practical methods to extract characteristic information, and each has its own peculiarity. Furthermore, the latter one may dispel some influences originated from the differences of instrument and operating environment. The study of two methods offers a new direction to the theoretic development and practical application of NIRS.