Novel Convolution-Based Signal Processing Techniques for an Artificial Olfactory Mucosa

Abstract

Abstract-As our understanding of the human olfactory system has grown, so has our ability to design artificial devices that mimic its functionality, so called electronic noses (e-noses). This has led to the development of a more sophisticated biomimetic system known as an artificial olfactory mucosa (e-mucosa) that comprises a large distributed sensor array and artificial mucous layer. In order to exploit fully this new architecture, new approaches are required to analyzing the rich data sets that it generates. In this paper, we propose a novel convolution based approach to processing signals from the e-mucosa. Computer simulations are performed to investigate the robustness of this approach when subjected to different real-world problems, such as sensor drift and noise. Our results demonstrate a promising ability to classify odors from poor sensor signals.

Applying Convolution-Based Processing Methods To A Dual-Channel, Large Array Artificial Olfactory Mucosa

“…The data obtained from the experiment were analysed using the methods reported in [8]. This processing approach takes related signals, such as a signal from before the retentive channels (S F ) and another from after passing through a retentive channel (S C ), and combines them into a new characteristic signal using the convolution transform (Equation 1).…”

“…One such method has been reported, where the spatio-temporal signals from matching spatially-separated sensors are combined and analysed using a convolution method [8]. This paper reports on the application of this novel processing approach to an artificial olfactory mucosa electronic nose.…”

Sample handling for electronic nose technology: State of the art and future trends

“…The idea lies in extraction of certain components from a mixture (that properly acts as sample enrichment) or temporal and/or spatial separation by analogy with functioning of biological nose [9,[82][83][84]. In the latter case, EN analyzes not the mixture as a whole but its individual components (or their groups).…”

“…Temporal and/or spatial separation of different mixture components makes it possible to obtain more adequate and representative information on the studied objects, especially if the initial ratios between the different components are not known [83,84]. It is well known that less volatile (high boiling) substances have higher energy of adsorption than high volatile (low boiling) ones [85].…”

Towards biological plausibility of electronic noses: A spiking neural network based approach for tea odour classification

“…This dichotomy and with the increased understanding of biological olfaction mechanisms, has led to the realization of several bio-inspired artificial olfactory systems not only at the abstraction level but also at the data processing stage [4,5]. Also, with the advent of novel bio-mimetic systems such as artificial olfactory mucosa [6][7][8] and an increase in the sensor array size it becomes challenging to be able to reduce the dimensionality while preserving relevant information in addition to computational overheads [7,8]. This calls for advanced time dependent bio-inspired computational techniques to address challenging odour discrimination problems like segmentation [8].…”