Study On The Quantitative Theophylline Analysis With Near-Infrared Spectroscopy Technique

Near Infrared Spectroscopy technology is a fast, efficient and non-intrusive analysis technology, which has been widely applied in pharmaceutical analysis field. It can not only be used for medicine in various physical states, such as materials, capsule and liquid preparations, but also for various kinds of medicines, such as protein, Chinese herbal medicine and antibiosis.Firstly, this paper introduced the creation of near infrared spectrum, its features, as well as the theoretic principles of the near infrared spectrum technology. Then, the feature and procedure of near infrared spectrum analysis is outlined, and the development of the near infrared spectrum technology is summed up. Following on, this paper brought forward the experiment scenario for the predicting model of the theophylline thickness, which employs the IRPrestige-21 Fourier Transformation Infrared Spectral Instrument to conduct full spectrum scan of the medicine sample, then to generate the absorbency graph, which is the base for building the thickness correction model. This research uses the BP artificial neural networks to build the correction model. In order to avoid the over fitting problem, the value of closeness is introduce to detect over fitting and thus optimize the model parameters. This paper studied the influence to the BP network, when conducting 1-order differential preprocess to the spectrum data, and the repetition and stability of this methods were verified. The test results show that the 1-order differential preprocess can efficiently reduce the noise in spectrum data, increase the accuracy of thickness prediction of any unknown sample. This paper compared the models build by conventional metrology methods such as Multiple Linear Regression Analysis, Principle Component Regression and Partial Least Square Regression, to models build by BP networks, and demonstrate the superiority of BP networks in solving non-linear classification. In the last, this paper pointed out several shortcomings of this research, and possible improvements.