A Hardware-Deployable Neuromorphic Solution for Encoding and Classification of Electronic Nose Data

Abstract

In several application domains, electronic nose systems employing conventional data processing approaches incur substantial power and computational costs and limitations, such as significant latency and poor accuracy for classification. Recent developments in spike-based bio-inspired approaches have delivered solutions for the highly accurate classification of multivariate sensor data with minimized computational and power requirements. Although these methods have addressed issues related to efficient data processing and classification accuracy, other areas, such as reducing the processing latency to support real-time application and deploying spike-based solutions on supported hardware, have yet to be studied in detail. Through this investigation, we proposed a spiking neural network (SNN)-based classifier, implemented in a chip-emulation-based development environment, that can be seamlessly deployed on a neuromorphic system-on-a-chip (NSoC). Under three different scenarios of increasing complexity, the SNN was determined to be able to classify real-valued sensor data with greater than 90% accuracy and with a maximum latency of 3 s on the software-based platform. Highlights of this work included the design and implementation of a novel encoder for artificial olfactory systems, implementation of unsupervised spike-timing-dependent plasticity (STDP) for learning, and a foundational study on early classification capability using the SNN-based classifier.

Bio-Inspired Strategies for Improving the Selectivity and Sensitivity of Artificial Noses: A Review

“…Pfeil et al [195] introduced supervised learning based on reward-depending plasticity with an accuracy ranging from 87% to 96% (Figure 11b). Recently, Vanarse et al [196] deployed a spiking neuron network hardware with a 96.5% accuracy in the identification of target gases. including onset latency for a better classification [194].…”

Application of a Brain-Inspired Spiking Neural Network Architecture to Odor Data Classification

“…Recent developments in neuromorphic olfaction have focused on leveraging the inherent advantages of the spike-based data representation to develop practical e-nose systems where key aspects such as data-to-spike encoding techniques, utilization of SNNs for pattern-recognition, and implementation of these models on low-power hardware are emphasized [ 17 , 18 , 19 , 20 , 21 , 22 ]. However, these neuromorphic models have mainly focused on data transformation based on biological spike encoding architectures, while overlooking the overall performance of the system to identify target odors with minimum computational resources and latency.…”

“…Although studies based on traditional machine learning methods for odor classification of the same dataset have claimed to have achieved high accuracy [19,34], these methods impose substantial computational and power requirements. Moreover, these techniques often require complex processing constructs and iterative training, resulting in considerable latency to provide recognition results, and the generalization capacity may also be limited [3,18]. Other neuromorphic approaches based on the same datasets have either focused on implementing data transformation based on the biological olfactory pathway or hardware-friendly application.…”

“…Hence, these approaches differ vastly in terms of encoding techniques and application of SNN for classification, thus making a direct comparison impractical. However, when evaluated against the spike-based approach described in [17][18][19], the 3D SNN model produces comparable results even when applied on original sensor responses without using any pre-processing or feature extraction. Moreover, a trained NeuCube model was able to provide a recognition result within a maximum processing latency of 950 ms on a medium specification desktop computer, inclusive of latencies resulting from software-based input/output and other programming constructs.…”

Optical Axons for Electro-Optical Neural Networks

“…The OA can be used in optical connections to the upper neural layers of mobile neuromorphic sensors for conversion of analogue inputs into the spiking frequency [40]. Analog-to-spike conversion has been used in the implementation of neuromorphic sensors in artificial olfactory systems [41][42][43] and in tactile sensors based on memristors [44]. Other neuromorphic sensors have been used for (i) control of the contraction force of shape memory alloy actuators [45]; and (ii) rotation control of robotic junctions [46].…”