Fig. 3: Deep-learning-based identification of mono-gas environment.
a Architecture of deep convolutional neural network (D-CNN) for classifying five gas species (air, methanol, ethanol, NO2, and acetone) and for quantifying the concentrations of each gas. b Classification accuracy and regression loss for the training and validation datasets with respect to the training iterations. c, d Gas species prediction results summarized in a confusion matrix and gas concentrations normalized to 0–1 for test dataset. The predicted gas concentrations are close to the identity line (y = x) with r2 = 0.888. e, f Real-time prediction of gas species and concentrations of methanol and ethanol. Here, normalized gas concentrations are converted to ppm by multiplying the maximum tested concentration of each gas
