Construction of a universal model for non-invasive identification of penicillins for injection using near-infrared diffuse reflectance spectroscopy

A universal NIR model for identification of 24 types of penicillins for injection has been developed. A total of 194 batches of 24 products from 87 manufacturers in China were used in the study. The classification model is a principal component analysis (PCA) based model consisting of a primary identification library with four sub-libraries. The spectral frequency regions used were 6000–6400 cm^?1 and 8400–8900 cm^?1 in the main library, 6000–6800 cm^?1 in sub-library 1, 4100–12,000 cm^?1 in sub-libraries 2 and 3, and 6200–6400 cm^?1 and 4700–5000 cm^?1 in sub-library 4. The data preprocessing method is the first derivative with nine-point smoothing followed by vector normalization. The distances between spectra were calculated using factors 2–5 for the primary identification library, factors 4–7 for sub-library 1, and factor 2 for sub-libraries 2–4. The specificity of the model was validated, and it had a correct identification rate of approximately 99%. This study has not only confirmed, but also improved the strategy described in our early report (Chong et al. (2009) 11]) to build such a library for the identification of different medicines by NIR.