Electronic Nose Sensor Array Optimization via Noise-Injected Gating Networks

Sensor redundancy remains a major bottleneck in electronic nose systems. To tackle this, we introduce an end-toend sensor array optimization (SAO) framework. By integrating a Gumbel noise-based gating strategy, our approach minimizes the bias from initial model weights and progressively forces the sensor weights toward binary states (0 or 1) during the training phase. Experimental results show that 13 -sensor array down to 5 sensors actually boosts classification accuracy from 83.59% to 89.69%. Compared to traditional SAO methods, this framework offers a highly effective route for designing compact, precisiondriven electronic nose configurations.