Estimation of Atmospheric PM2.5 based on Photos and Deep Learning

Abstract

Particulate matter with a diameter of less than 2.5 microns (PM2.5) in the air is one of the most critical pollutants related to air quality. Exposure to high levels of PM2.5, which can be inhaled and carry harmful chemicals deep into the lungs and bloodstream, can have acute or chronic adverse health effects. A reliable, convenient and low-cost method to obtain PM2.5 concentration can help people improve their awareness of it. At the same time, it can also provide a certain reference for the prevention and control of haze, for air purification and for some other work to reduce the harm of air pollution to human health. In this paper, from the perspective of deep learning, with the help of photos that can be taken anywhere, we combined convolution neural network and support vector regression machine to estimate PM2.5. The method used in this paper is compared with other methods using the same dataset to prove its effectiveness.