초분광 반사 영상과 부분최소제곱회귀 모델을 이용한 우유 분말에 혼합된 미량 멜라민의 함량 예측

Melamine has been reported to be responsible for kidney stones and renal failure among infants and children. Conventional detection methods, High-Performance Liquid Chromatography (HPLC) and Gas Chromatography (GC), are sensitive enough to detect trace amounts of the contaminant, but they are time consuming, expensive, and laborintensive. Hyperspectral imaging methods, which combine spectroscopy and imaging, can provide rapid and nondestructive means to assess the quality and safety of agricultural products. In this study, near-infrared hyperspectral reflectance imaging combined with partial least square regression analysis was used to predict melamine particle concentration in dry milk powder. Melamine particles, with concentration levels ranging from 0.02% to 1% by weight ratio (g/g), were mixed with dry milk powder and used for the experiment. Hyperspectral reflectance images in the wavelength range from 992.0 nm to 1682.1 nm were acquired for the mixtures. Then PLSR models were developed with several preprocessing methods. Optimal wavelength bands were selected from 1454.5 nm to 1555.6 nm using beta-coefficients from the PLSR model. The best PLSR result for predicting melamine concentration in milk powder was obtained using a 1^st order derivative pretreatment with ${\text{R}}_{\text{v}}^2=0.974$, SEP=±0.055%, and F=6.