A Brain-Inspired Hybrid Plasticity Feedforward-Inhibitory Spike Neural Network for Olfactory Perception

Brain-inspired spiking neural networks (SNNs), which mimic the information processing mechanisms of the biological mammalian olfactory bulb, offer enhanced biological interpretability, more efficient extraction of spatial features from gas sensor array data, and superior energy efficiency. These advantages endow electronic nose (e-nose) systems with promising application potential in complex olfactory scenarios. However, in gas qualitative identification and abnormal gas detection tasks, SNN models inspired by mammalian olfactory bulb neurons face two primary limitations. First, they lack a hybrid plasticity (HP) learning mechanism that integrates correlation-driven and error-driven processes. Second, they lack efficient spike encoding strategies tailored to the response characteristics of gas sensor arrays. Therefore, this article proposes a brain-inspired HP feedforward-inhibitory SNN (FI-SNN) model. By designing an accumulate rate spike encoder, constructing a feedforward-inhibitory spiking network structure, and introducing an HP learning method, the model enhances the spatial feature representation capability of spike encoding for gas sensor array data, enables HP learning in olfactory perception SNN models, and thereby significantly improves the performance of gas detection. Experimental results demonstrate that the proposed model achieves competitive accuracy, reaching 99.48% on the VOC gas detection dataset and 97.96% on the red wine quality monitoring dataset, surpassing the state-of-the-art methods.