Brand classification of detergent powder using near-infrared spectroscopy and extreme learning machines

Abstract

Detergent powder is a necessity in daily life and has a wide range of sales worldwide. To obtain illegal benefits, immoral merchants often sell inferior detergent powder labeled as famous brands. Thus, there is a pressing need for developing convenient methods able to identify different brands of detergent powder. In this study, near-infrared (NIR) spectroscopy is applied for the identification of 6 different brands of detergent powder. Both extreme learning machine (ELM) and its ensemble (EELM) are used for constructing predictive models. A total 180 samples belonged to 6 different brands are prepared for experiment. A simple filter algorithm, i.e., Relief, was applied to select informative features/variables. Principal component analysis (PCA) is used for preliminary exploratory analysis. All samples were divided into subsets with equal size for training, optimizing and testing the predictive model, respectively. The model from the EELM algorithm achieved 100% accuracy on both the training and test sets, which is superior to the model from the ELM algorithm. It indicates that NIR spectroscopy combined with feature compression and the EELM algorithm is a potential tool for identifying the brands of detergent powder..