Abstract
Electronic nose systems have been used to detect odorous molecules in industrial and environmental applications. Most existing algorithms for such applications are based on the concept of nearest-neighbor classification which computes the "distance" between the test odor and a set of known odors (also called training odors), and may erroneously classify an unknown odor as an odor in the training data set. In this thesis, we propose to improve the situation by treating the unknown odor (if not belonging to the training data set) as an unclassified odor, and use an outlier rejection mechanism to avoid misjudgment. We also designed and fabricated an integrated system chip (containing sensor interface circuitry, analog-to-digital converter, 8-bit microcontroller, and memory) for use in a portable electronic nose system. Experimental results are shown to justify the effectiveness of our proposed algorithm and integrated chip.