Identification of Gas Type Using Thermal Camera and Convolutional Neural Network

Gas content measurements are frequently employed to detect hazardous levels of air contamination that pose health risks. Additionally, leak detection at fuel oil filling stations and natural gas distribution installations is crucial to prevent potential fires and explosions. In this study, a gas type identification system was developed, utilizing a thermal camera and a neural network. The thermal camera captures infrared intensity, comprising 768 pixels organized in a 32×24 image matrix, represented with a monochromatic color scale. Equipped with two optical filters with wavelengths of 5.64 and 4.63 μm, this camera records distinct intensity patterns absorbed by the sample gas for each filter. The tested air samples encompassed butane, cigarette smoke, alcohol, gasoline, ammonia, vehicle exhaust gases, and clean air. The classification of gas types was achieved through a 1D Convolutional Neural Network (CNN) deep learning algorithm. The test results demonstrate that this system effectively detects and distinguishes each gas type with an accuracy level exceeding 91%.