Author Identifiers

Anup Vanarse
ORCID: 0000-0002-0082-4283

Date of Award


Degree Type


Degree Name

Doctor of Philosophy


School of Engineering

First Advisor

Associate Professor Adam Osseiran

Second Advisor

Associate Professor Alexander Rassau


Electronic nose systems, popularly known as e-noses, are one of the classic examples of analytical devices that have been researched extensively, but have had limited commercial success for applications outside of a laboratory environment. Based on the idea of emulating the biological olfactory pathway, e-nose systems generally consist of a chemo-resistive array as a sensing front-end that transduces the interaction with aromatic compounds into electrical signals. In the next stage, a signal conditioning unit performs pre-processing and feature extraction, and modulates the sensor responses into unique “odour-prints” to represent a chemical compound. Finally, a pattern-recognition engine is implemented that provides odour identification results. While this three-stage architecture seems simplistic, the realisation of each stage is significantly complex, starting from the selection of appropriate sensing materials for the front-end array to the handling of the highly multi-dimensional data generated, and the implementation of effective pattern-recognition algorithms for this data.

Although advances in computing techniques have enabled a variety of algorithms for preprocessing, feature extraction, and pattern-recognition, their short-comings in terms of computational resource requirement, processing latency, and classification accuracy have largely limited the application of e-nose systems to laboratory environments. Moving away from statistical pattern-matching techniques, e-nose systems greatly benefited from application of conventional machine learning approaches for generation of meaningful features, application of dimensionality reduction, and more advanced pattern-recognition techniques. However, these improvements were insufficient to overcome the effects of their data-intensive structure and implementation complexity that hindered their performance in real-world applications.

The emergence of neuromorphic engineering, a bio-inspired method that mimics the neuro-biological architecture by encoding and processing information using sparse spike-based representation with minimal power consumption, delivered promising results for vision and auditory sensors. With increasing interest in developing low-power and robust chemical sensors, the application of neuromorphic engineering concepts for electronic noses has provided an impetus for research focusing on improving these instruments.

While conventional e-nose systems apply computationally-expensive and power-consuming data processing strategies, neuromorphic olfactory systems implement the biological olfaction principles found in humans and insects simplifying the handling of multivariate sensory data by generating sparse spike-based information and applying spiking neural networks for processing. Over the last decade, research on neuromorphic olfaction has demonstrated the capability of these systems to tackle issues related to classification accuracy, processing latency, power consumption, and size-reduction that plague current e-nose implementations. Despite the progress in this field, what is lacking is a neuromorphic olfaction system that is both robust and able to provide real-time identification of volatile compounds in a practical application.

The main objective of this research has been to develop a neuromorphic pattern-recognition engine for an e-nose system that can function in real-time and provide reliable identification information. To achieve this goal, we apply and validate different spiking neural network architectures with hardware or cloud-based deployability to inform the design of models that can be applied in a real-world setting. In the process of developing neuromorphic classifiers, we also apply different data-to-spike encoding approaches to evaluate their efficiency and effectiveness to encode multivariate data in a sparse representation.

Significant contributions from this investigation include the development of neuromorphic models that provide highly accurate classification results with minimum latency, the development of a novel data encoding tool based on a de-facto standard bus used in neuromorphic systems, demonstration of the integration of a neuromorphic pattern-recognition model with a commercial e-nose system through a real-world application, and preliminary results showing the applicability of the proposed neuromorphic models in e-tongue systems for the development of neuromorphic gustatory systems. Overall, through this work, several novel approaches for implementation of a neuromorphic framework as a pattern recognition engine for rapid and reliable classification of continuous multi-variate e-nose data are presented.

Available for download on Sunday, October 12, 2025