Abstract: In this approach, the intelligent portable electronic nose with a gas sensor array containing ten gas sensors was used to collect the odor information of different varieties of pungent Chinese herbals. In order to get a new sensor data processing solution to pattern recognition of the Chinese herbals accurately, a manifold algorithm named Locally Linear Embedding（LLE）was used to finish the dimensionality reduction and feature extraction of the odor data. And then the Fisher Linear Discriminant Analysis(LDA) was adopted to complete the identification via clustering. After many times of experiments to optimize the parameters of LLE, the classification results are quite perfect. The results demonstrate that by using this new combined method, the electronic nose can not only distinguish 6 types of Chinese herbals, but also identify the 3 different batches and the 3 different denominations of origin accurately, which provides a new method for further analysis.

Key words: pattern recognition, gas sensors, locally linear embedding_linear discriminant analysis（LLE_LDA）, classification, nonlinear dimensionality reduction