Abstract: An ultra-fast gas chromatography-based electronic nose was used to qualitatively and quantitatively analyze the volatile flavor components in 36 samples of whole milk, low-fat milk, and skimmed milk subjected to ultra-high temperature (UHT) treatment. Using a combination of principal component analysis (PCA) and multi-layer perceptron neural network (MLP-NN), UHT-treated milk samples were classified and predicted according to their aroma characteristics. The results showed that 29 volatile flavor components were detected in all milk samples, mainly including acetone, n-butanol, and δ-decalactone, acetone and decanal being the most and least abundant, respectively. Isobutanol, acetoin, 1-pentanol, E-3-hexenal and decanal only existed in UHT-treated whole milk. 3-Methylheptane, 2,6-dimethylpyrazine, and E-2-nonen-1-ol were exclusively detected in UHT-treated low-fat milk. α-Pinene, 5-methylfurfural and decanoic acid were only detected in UHT-treated skimmed milk. The major aroma characteristics of UHT-treated milk were vegetable-like, creamy, grassy, fruity, and malty aroma notes. Among them, vegetable-like, creamy and malty aromas were the most widely distributed in UHT-treated whole fat milk. Fruity aroma was the most widely distributed in UHT-treated low-fat milk. Grassy aroma was the most widely distributed in UHT-treated skimmed milk. PCA showed that there was a significant difference between different brands of UHT-treated milk. The prediction accuracy of MLP-NN for UHT-treated milk was as high as 98.6%.

Key words: ultra-high temperature milk; ultra-fast gas chromatography-based electronic nose; volatile components; principal component analysis; multi-layer perceptron neural network