Abstract: A support vector machine (SVM) model for rapidly predicting the total volatile basic nitrogen (TVB-N) content in prawns was established using near-infrared spectroscopy and machine vision technology based on spectral and image information infusion. Spectral and image information of 51 samples stored at 4 ℃ for 0 to 12 days was obtained, and their TVB-N content was measured according to the national standard method of China. The results showed that the characteristic dual-band spectral information in the range of 350–1 000 nm and 940–1 650 nm was fused and pre-processed by the first derivative method. Besides, the competitive adaptive reweighted sampling (CARS) algorithm was used to select the characteristic wavelengths. The developed model worked well with correlation coefficient in the prediction set (Rp) of 0.968 7, standard error of prediction (SEP) of the validation set of 10.56 mg/100 g, and relative percent deviation (RPD) of 3.38. On the other hand, the model constructed using the characteristic image information had poor performance, with Rp of 0.933 5, SEP of 19.79 mg/100 g and RPD of 1.74. Compared with the above two models, the model developed based on spectral and image information fusion had improved accuracy and stability with Rp of 0.988 4, SEP of 7.51 mg/100 g and RPD of 6.29. These results confirmed the potential near-infrared spectroscopy combined with machine vision to predict the TVB-N content in prawns, which could allow rapid detection of freshness changes of prawns during cold storage.

Key words: near-infrared spectroscopy; machine vision; volatile basic nitrogen content; support vector machine model