Abstract: We developed a near-infrared (NIR) spectroscopic method for rapidly identifying the adulteration of lotus seed flour. The spectral data of lotus seed flours mixed with different other crop flours at different levels were classified using support vector machine (SVM). On this basis, deep belief network (DBN) was employed for discrimination of lotus seed flours with unknown adulterants. The results showed that the average recognition rate of the SVM model for adulteration of the other crop flours was 98% (training numbers ≥ 600). Based on the DBN model, lotus seed flour adultered with the other crops could be effectively identified and the average recognition rate for very few adulteration levels was about 96% (training numbers ≥ 600). The DBN model avoided the disadvantages of the current deep neural networks such as local optimization and not using massive labelled samples. NIRs combined with DBN provides a new alternative for the rapid detection of adulteration of agricultural products.

Key words: lotus seed flour, adulteration, near-infrared spectroscopy, support vector machine, deep belief network