Mammalian Olfaction-Inspired Spike-Coded Bionic Neural Network

To overcome the complex signal processing challenges in electronic nose (e-nose) odor detection, this study proposed a biomimetic olfactory perception neural network (BOPNN), inspired by mammalian olfactory systems. The BOPNN, mimicking mammalian olfaction structure, directly analyzes sensor outputs, using Gabor receptive field-based neural coding scheme to bypass traditional preprocessing. To assess the effectiveness of the proposed bionic model, comparative analyses were performed with several classic classification methods and various bionic models that were previously developed by other researchers. Experimental results revealed that the BOPNN achieves an average accuracy of 94.6% on Dataset A (single odor), surpassing the 87.5% recognition rate of classic methods. Additionally, the model was tested on Dataset B (mixed odors), with an average of 85% higher than other comparative methods demonstrating its robustness in complex environments. Furthermore, scenarios with damaged sensors in the array were simulated. Experiments demonstrated that the model is robust to variations in sensor type and quantity. Notably, even when more than 50% of the sensors in the array were damaged or unresponsive, the model maintained a recognition rate of over 85% on Dataset A.