A robust and fast odor detection method with remaining response curve forecasting

Pattern recognition constitutes a critical building block in machine olfaction. Traditional machine learning methods rely on the response curves from gas sensor arrays. However, it usually takes hundreds of seconds to acquire the entire curves, which significantly increases detection time. Furthermore, manual feature extraction suffers from limitations such as low adaptability and inefficiency. Therefore, a robust and fast gas detection method combining Remaining Response Curve Forecasting (RRCF) and Gas Classification (GC) is proposed. First, the early part of the response curve is used as input, and RRCF predicts the remaining part of the curve. Second, the early part and prediction result are spliced together to reconstruct the response curve. Finally, GC utilizes these reconstructed response curves for gas classification. Experimental results show that RRCF-GC achieves accuracies of 97.58 % and 91.59 % in cross-board and mixed gas scenarios, respectively, under the condition of reducing the detection time by about three-fifths.