Abstract: Near-infrared spectroscopy combined with chemometric methods was employed to determine the acid value of third grade soybean oil as an improvement over the current methods. The characteristic waveband of 4500－6000 cm-1 was selected to establish an acid value calibration model. Spectral pre-processing methods based on wavelet transform were investigated and the effects of different wavelet decomposition levels on prediction results of acid value were compared in detail. The results showed that Daubechies4 (db4) wavelets with four levels of decomposition revealed the best noise removal. A back propagation (BP) neural network calibration model was built based on reconstructed spectral signals with wave filtrating and was validated using the prediction set with a determination coefficient of 0.9743 and a RMSEP of 0.1036. This study supports a high feasibility to apply near-infrared spectroscopy for the rapid determination of acid value of oils.

Key words: acid value detection in oil, near infrared spectroscopy analysis, wavelet pre-processing, partial least square regression, BP neural network