Abstract: An electronic nose was designed to collect odor data of live Chinese mitten crab (Eriocheir sinensis) using a sensor array consisting of 7 commercial tin oxide gas sensors. To obtain a better feature vector for identifying different crabs, a modified unsupervised discriminant projection coupled with sample label information was proposed which could maintain the local and global structure and take advantage of the important label information to achieve optimal linear geometric projection. Then back-propagation neural network was used for modeling the quality changes of crabs during storage. At the same time, the total volatile basic nitrogen (TVB-N) of crab meat was measured and used as an indicator of crab freshness. The results showed that a high degree of accuracy in nondestructive identification of crab freshness was achieved with electronic nose based on this algorithm.

Key words: Chinese mitten crab, freshness, electronic nose, unsupervised discriminant projection algorithm, back-propagation neural network