Deep Gated Channel Selection Mechanism for Sensor Array Optimization in Electronic Nose

Sensor array optimization (SAO) seeks to identify the minimal subset of sensors that preserves maximal information for odor analysis, thereby reducing the cost and complexity of electronic noses (e-noses). This work presents a novel SAO method based on a gated channel selection network, termed the sensor selector. The sensor selector learns a soft “0-1” mask by capturing nonlinear relationships both within and across sensor responses, adaptively retaining informative signals while suppressing redundant ones. Unlike conventional step-by-step SAO methods, which handle feature extraction, feature selection, and sensor selection separately, the proposed sensor selector enables an end-to-end SAO workflow by integrating the feature extractor and task-specific model into a unified learning framework. Experiments on a 13-sensor array for the classification of eight target odors demonstrate that our method achieves superior performance with five sensors, highlighting the potential of gated sensor selection to improve both efficiency and accuracy in practical e-nose applications.