Video classification method of key pollution sources based on deep learning

Four researchers based in Guangdong, China created a method of applying artificial intelligence to rampant air, water, and land pollution. In their 2019 study, they highlight how China’s swift economic growth and industrialization has severely strained and damaged the environment, along with the public’s growing discontent with the state of affairs. However, with so many different sources of pollution and ineffective ways to trace and measure pollution as a whole from multiple sources, trying to tamp down pollution is incredibly challenging. Using deep learning in a convolutional neural network (CNN), a form of artificial intelligence, the researchers have essentially found an automated method of classifying, quantifying, and possibly even identifying early signs of a pollution event (such as illegal waste dumping) through processing video footage and classifying the shade and texture of the footage pixels. The researchers trained the CNN using 6,000 video samples of primarily water pollution events occurring and not occurring, and through those samples were able to essentially establish a methodology for the CNN to correctly identify and classify pollution 93% of the time. In application, the CNN would require broad use of cameras, which may not always be practical, and some sources of pollution, like air pollution, are harder to capture and classify, but this research does present great potential for authorities to be alerted of the type and initial location of a pollution event, initiating site testing and regulation enforcement faster and more effectively.


Artificial Intelligence, Ecological Monitoring, Pollution, Regulation, Visual Technologies