A CNN-based simplified data processing method for electronic noses

Traditional data processing methods for electronic noses (e-noses) need to use the whole response curves (including rise, steady and recovery phases) of sensor array, which leads to a long sampling time. The traditional methods also perform many steps such as signal pre-processing, feature generation/reduction, and classification, which increase the difficulty of selecting a suitable method for each step. In view of the above problems, we present a simplified method based on convolutional neural network (CNN) for e-noses. CNN not only uses fewer sampling points to perform classification, but also can automatically implement feature generation without signal pre-processing step. This significantly improves detection efficiency and simplifies data processing procedures of e-noses. The performance of CNN was tested using a portable e-nose designed by us. The results showed that CNN not only took shorter sampling time (15 seconds), but also obtained higher classification accuracy (95.7%) than traditional methods (92.9%).