The Recognition of Different Odor Using Convolutional Neural Networks Extracted From Time and Temperature Features

In this paper, we propose a dataset processing method for odor recognition using neural networks, which extracts the sensor multi-dimensional temperature information and high-dimensional time series information during the response process. In this “sniffing” system, 4 different Micro-Electro-Mechanical System (MEMS) Metal-Oxide-Semiconductor (MOX) gas sensors are utilized, and 4 different heating voltages are applied to the sensor array. After classifying 6 chemical gases closely related to food at 3 different concentrations each, we introduced 10 kinds of food for classification. To construct an abstract odor factor map as a feature map of odor samples, we process each group of different relatively low-bandwidth sensor signals with smoothing spline, Gaussian window, etc. By Inputting the feature map into a specific convolutional neural network, we successfully achieve 98.7% accuracy in chemical gas recognition and 95.8% accuracy in food classification.