Development of an Electronic Nose for Olfactory System Modelling using Artificial Neural Network

• Main Authors: Fernandez, Proceso L, Jr, Roa, Mary Anne Sy
• Format: text
• Published: Archīum Ateneo 2018
• Subjects: Artificial Neural Network; Odor Classification; Electronic Nose; Machine Learning; Artificial Intelligence and Robotics; Computer Sciences
• Online Access: https://archium.ateneo.edu/discs-faculty-pubs/74; https://archium.ateneo.edu/cgi/viewcontent.cgi?article=1073&context=discs-faculty-pubs
• Institution: Ateneo De Manila University

Electronic nose (e-nose) devices have received considerable attention in the field of sensor technology because of their many potential uses such as in identification of toxic wastes, monitoring air quality, examining odors in infected wounds and in inspection of food. Notwithstanding the vast amount of literature on the usage of e-noses for specific purposes, the technology originally and ultimately aims to mimic the capability of mammals to discriminate odors from all sorts of objects. This study demonstrates the theoretical and practical feasibility of designing an e-nose towards general odor classification. A multi-sensor array hardware unit was carefully constructed for data collection and odor detection. Important hardware design considerations such as sensor calibration, aeration, circuit protection, and voltage/current requirements were satisfied. A highly fine-tuned artificial neural network (ANN) was integrated to the hardware to interpret and relate the data to a target odor class from a set of 10 primary odors identified in a previous study. Various network architecture considerations, such as neuron count, number of layers and activation function, as well as various data treatment methods, such as normalization, and data partitioning, were investigated. The results showed that careful hardware integration with an ANN having sufficiently deep internal structure can yield accurate classification to at least half of the ten primary odor classes, namely fragrant (96%), fruity (98%), chemical (99%), peppermint (98%), and popcorn (90%). The results demonstrate the feasibility of making e-noses for general odor classification, which could lead to further broadening of e-nose applications.