Abstract: The aroma profiles of wolfberry were studied by the electronic nose (E-nose) for aim of subjective and fast discrimination of the geographical origin of wolfberry. The effects of sample weight, headspace-generated time, and headspace volume on sensor responses were studied by single-factor experiments. Results of one-way analysis of variance found that the responses of E-nose sensors were significantly affected by these factors. The sample weight showed significant effect on S7 while very significant effect on the other 9 sensors. The effects of headspace-generated time were very significant on the sensors except for S2, S7, S9 and S10. With the help of canonical discriminant analysis (CDA), the optimum experimental parameters were acquired: flow rate of 300 mL·min^-1, 20 g of sample sealed in 500 mL beaker for 30 min headspace-generated time. With the optimum experimental parameters, samples produced in three different regions (Guazhou Gansu, Chaidamu Qinghai, Zhongning Ningxia) were detected. With PCA and CDA, the wolfberries were grouped according to the geographical origin, with three samples from Zhongning overlapped with each other. BPNN were employed to build the predictive model for the geographical origin of the wolfberry fruit samples, with 96% samples correctly predicted. The E-nose was proved to be useful for the identification of geographical origin of the wolfberry samples for its efficiency and high accuracy, which laid solid foundation for the traceability of wolfberry geographical origin.

Key words: wolfberry, electronic nose, multivariate data analysis, geographical origin