ASASA: Automatic Selection and Adaption in Sensor Array for Intelligent Artificial Olfaction Systems

The aim of developing artificial olfaction systems (AOSs) was to mimic the biological olfactory sense with the help of electronic odor/gas sensors. Such systems may show potential in all the domains wherever the sense of olfaction plays a role. In several application areas, a modular architecture of the traditional AOS is used (i.e., separate modules for preprocessing and decision making). On the other hand, such systems are expected with rapid outcomes in real-time end-to-end applications while consuming optimal resources. Hence, the modular architecture and expected performance show a contradiction. This letter proposes "automatic selection and adaption in sensor array (ASASA)" to curb this issue. It facilitates automatic preprocessing that also makes the data compatible with the classifiers for direct usage in an end-to-end manner. The proposed methodology makes the traditional AOS intelligent. The obtained experimental results on three different datasets show the effectiveness of ASASA. Moreover, the classification accuracy using tiny convolutional neural network achieved a score of more than 93.50% for the considered datasets.