Electronic nose, HS-GC-IMS, HS-SPME-GC-MS, and deep learning model were used to analyze and predict the changes and contents of VOCs in in-shell walnut kernels under different roasting conditions.

Roasting imparts walnuts with an increased amount of hedonic aromas. Consequently, this study comprehensively analyzed aroma characteristics of in-shell walnut kernels during roasting at different conditions using quantitative descriptive analysis, Electronic-nose, HS-SPME-GC-MS, and HS-GC-IMS. The results indicated that roasting at 140 °C for 60 min was optimal for enhancing aromatic compound release. Seventy-six volatile organic compounds (VOCs) were detected using HS-SPME-GC-MS, while HS-GC-IMS identified 47 VOCs. Aldehydes, ketones, alcohols and esters were the main VOCs in walnut (80 %). Among them, 1-Propanethiol, Ethanol, 2-Butanone, Butanal, and 1-Penten-3-one were used as marker VOCs at different roasting times. 2-Methylpropanal, 1-Hexanal, Cyclopentanone, and 1-Pentanol as marker VOCs at different roasting temperatures. Finally, the backpropagation neural network model showed satisfactory results in predicting the VOCs content (accuracy rate 0.9448). The findings offer an important understanding regarding the development of flavor in whole walnut fruits while roasting, as well as the improvement of the roasting procedure.