Combining traditional and spiking neural networks for energy-efficient detection of Eimeria parasites

Abstract

The detection of bacterial and viral microbes is pivotal for both human and animal well-being in the public health services and for veterinary care. Even in a laboratory, the isolation of microorganisms requires time-consuming procedures and expert technicians. However, the rise of Machine Learning and Deep Learning has seen a surge in the application of techniques that can be applied outside the laboratory to classify microorganisms using microscopy images. Yet, despite their success in various domains, Deep Learning approaches tend to have high energy demands, which can in some contexts limit their application, increasing both costs and environmental concerns. In this study, a novel hybrid methodology was proposed in which an Artificial Neural Network was combined with a Spiking Neural Network. A quality measure was proposed to assess the effectiveness of the hybrid models, in which their energy efficiency, energy consumption patterns, performance levels and accuracy were all considered. The synergy of both methods markedly reduced the energy footprint of deep-learning models programmed to detect microorganisms, increasing their environmental sustainability and the feasibility of their use in places with little or no electricity supply. The efficacy of our model was demonstrated through the detection and the classification of different species of the Eimeria parasite on chicken and rabbit farms.