Using a Hybrid Deep Neural Network for Gas Classification

In terms of electronic nose algorithms, data pre-processing and classifier type are the two main factors affecting gas classification results. In the early stage, data pre-processing mostly takes specific information from gas-reaction waveforms as features and uses machine learning algorithms, such as K-Nearest Neighbor(KNN) and Support Vector Machine(SVM), to classify the gas data. In recent years, some research has been done on using deep learning for gas classification. The data pre-processing takes the overall process of the gas reaction as a feature map, and the classifier uses Convolutional Neural Network(CNN) architecture to classify the gases, resulting in classification accuracy being significantly higher than those of traditional machine learning algorithms. In addition, external factors such as wind speed, and distance from the gas source are also important factors affecting gas classification. The objectives of this study are as follows: 1) improving the data pre-processing method and classifier structure in deep learning for gas analysis and 2) using hybrid deep neural networks with Multilayer Perceptron (MLP) for environment compensation to improve the sensor drift problem caused by external factors. This study used one open-source gas dataset, applied three data pre-processing methods and two deep learning architectures (GasNet, SimResNet-9) for gas analysis and comparison, selected the method with the best classification accuracy and used it in Deep Neural Networks with MLP environmental compensation to promote the accuracy of classification further by learning the relationship between external factors and gas data. The proposed SimResNet-10_X_MLP was used for data training and classification in this study, achieving a 95% classification accuracy.