Study on an Improved LeNet-5 Gas Identification Structure for Electronic Noses

This paper presents a new convolutional neural network structure for gas identification of electronic noses. Inspired by the tremendous achievements that made by convolutional neural network in the field of computer vision, based on the typical LeNet-5 structure, an improved LeNet-5 is designed for a 12 sensor array to identify three categories of CO, CH4 and their mixtures omitting the concentration influences. Experimental results show that the final gas identification accuracy rate reaches 99.67% with the improved LeNet-5. Comparison with results of Multiple Layer Perceptron neural networks also verifies the effectiveness of the proposed approach.