Establishing the regression model of smoke-flow opacity for adaptive intelligent smoke identification using remote hyperspectral sensing

Abstract

Industrial emissions are one of the major sources of pollution in urban areas. Air pollution due to industrial emissions has been a persistent and challenging issue ever since the Industrial Revolution began several centuries ago. Air pollutants refer to smoke and suspended particles emitted during fuel combustion processes such as transportation, industrial production, and energy generation. These air pollutants usually have a significant impact on atmospheric visibility. Serious air pollution has a profound impact on human health. Its concentration is positively correlated with the incidence and mortality of upper respiratory system, lung, cardiovascular, and other diseases. Complied with the 2050 Net-Zero global goal, how to control and detect air pollution smog has become an important and essential issue for advanced countries. In 1993, the United States implemented the visual smoke identification method for visible pollution sources (Visible Emission Field Manual EPA Methods 9), and established training guidelines for manual visual smoke identification. However, it is difficult to have objective standards for visual judgment of smoke, and it is also affected by sunlight, sky background, and smoke color. Consequently, in 2016, the United States implemented the Standard Test Method for Determining the Opacity of a Plume in the Outdoor Ambient Atmosphere (ASTM D7520-16). The screening of environmental restrictions in this method is the same as Method 9, which is determined by the certificated inspector. The smoke flow is photographed with a digital camera, and the digital image is then used to calculate the opacity of the smoke flow. ASTM D7520-16 only smoke opacity is captured and calculated using a digital system, and the judgment of environmental conditions is still judged strongly by the inspector. To improve the shortcomings of traditional inaccurate quantification and only specific air pollutants, this paper proposes to combine image recognition technology and deep learning analysis associated with the remote hyperspectral camera to determine the background of the smoke, sunshine, and atmospheric environmental conditions adaptively, thereby improving the accuracy of identifying smoke concentration intelligently. It also attempts to establish a regression model of various color smoke-flow opacity to increase the feasibility of smart smoke identification in the future.