Identification of industrial gas by sparse infrared absorption spectrum characteristics and support vector machine

Abstract

Most industrial gases such as methane(CH4), ethylene (C2H4) and sulfur hexafluoride (SF6) have obvious absorption characteristics in the infrared band. The infrared absorption spectrum of leaking gas can be obtained through multispectral or hyper-spectral detection technologies to realize gas identification. However, these methods need a lot of work calibrating the detector response curve to target gas. In this work, a sparse infrared absorption spectrum based support vector machine (SVM) recognition method is proposed to obtain the gas absorption peak information without response curve calibration. An uncooled infrared imaging component is utilized to compose a multi-broadband long-pass differential filter infrared imaging setup that filters in the range of 7.5μm ~13.5 μm. Data extracted from multi-band infrared images of C2H4 and SF6 collected by the setup, combined with the simulated data generated by the simulated sparse spectrum algorithm, constitute training set to SVM. C2H4 and SF6 can be accurately identified under laboratory conditions with the path-concentration of 500 ppm·m ~1000 ppm·m. The easy to implement and cost-effective method is expected to realize real-time identification of leaking gas.