Enhanced Gas Recognition of Electronic Nose Using 1-D Convolutional Neural Network With Savitzky–Golay Filter

The rapid development of signal processing technology has improved the stability and anti-interference ability of gas sensors in electronic noses (E-noses). However, the interference and noise caused by temperature and humidity in the environment are still inevitable in real detection conditions, which can cause data fluctuation during the recognition process. In traditional pattern recognition, the data fluctuation would reduce the difference between extracted features and the accuracy of gas classification. This study proposes a 1-D convolutional neural network (1DCNN) with the Savitzky–Golay (SG) filter. The SG filter is added before the convolution layer in the 1DCNN to automatically remove the noise of the sensor array data. The model can improve the effectiveness of the convolution layer to obtain features, omit the tedious preprocessing steps, and directly complete the identification process from raw data to results. The 1DCNN with SG filter is employed for the recognition of four gases: methane, ethanol, ethylene, and carbon monoxide. The results show that the accuracy of the 1DCNN with SG filter (99.21%) is 4% higher than that of the CNN (95.31%). Furthermore, the 1DCNN with SG filter is utilized for classifying a diverse assortment of mixed gases, ultimately achieving a classification accuracy of 99.8%. This study demonstrates the effectiveness of a novel CNN model with an SG filter, which streamlines the data processing and significantly improves the accuracy and efficiency of gas recognition in E-noses.