Temperature Limit Series Of Related Network Applications In The Near-infrared Spectral Data Analysis

The basic theory of temperature-constrained cascade correlation networks (TCCCNs) and its application in near infrared diffuse reflectance spectra were illustrated in this thesis. The powder pharmaceutical samples of sulfaguanidine are analyzed nondestructively by original near-infrared diffuse reflectance spectra and their first derivative spectra. The TCCCN method was applied to the classification of Sulfaguanidine based on near infrared reflectance spectra. And the effects of parameters and related problems were discussed. NIR spectra have the features of broader absorptive bands and weaker intensities. The spectra of different components are seriously overlapped in this region, peak separation was improved by using first derivative spectra. NIR spectra can be useful if only proper data processing methods are used and combined with chemometrics methods. In this thesis a temperature-constrained cascade correlation network (TCCCN) has been proved to be a good approach for the classification of qualified, un-qualified, and counterfeit pharmaceutical powder samples of sulfaguanidine. The TCCCN models were verified with independent prediction samples by using the "cross-validation" method. The results showed that single outputs network generally performed better than the multiple outputs networks, and the first derivative spectra were more suitable for the identification comparing with original diffuse reflectance spectra. The classification can be reached up to 100% when original reflectance spectra and Uni-TCCCN were used, and the classification can be only reached up to 95 % when original reflectance spectra and Multi-TCCCN were used. The classification can be both reached up to 100% when first derivative reflectance spectra and Uni-TCCCN, Multi-TCCCN were used, respectively. But when the classification is reached up to 100 %, the pre-defined relative error with Multi-TCCCN was smaller than that with Uni-TCCCN.