AI to detect sources of air pollution

This research, a collaboration with Lahore University of Mangement Sciences in Pakistan, uses AI to detect unregulated brick kilns, which are significant sources of air pollution and modern-day slavery in South Asia. By combining remote sensing data with high-resolution imagery, this model can identify these kilns more efficiently than previous models. 

Industrial air pollution, particularly from small-scale industries like brick kilns, poses severe health risks and contributes significantly to climate change. The South Asian "Brick-Kiln-Belt," is a major source of carbon emissions and modern-day slavery. Mapping and monitoring these kilns are crucial for addressing their detrimental effects. Traditional methods, such as manual surveys and crowd-sourced detection, are not scalable. Previous automated methods either lacked accuracy or were computationally expensive.

Our models uses a two-step method to detect kilns, using low-resolution spatio-temporal multi-spectral data from the Sentinel-2 satellite to use measures like moisture levels, built-up features, vegetation levels or burn scars to identify potential locations. These measures are shown in (i) combined into a heat map, where red represents potential kilns. This step allows the more precise but resource-intensive high-resolution detection using YOLOv3, a deep learning AI object detector, to analyse only the remaining data (ii) and rule out any false positives – this represents over 99% reduction in data volume.

By combining the low- and high-resolution data in a two-step process, this new AI-based method is 21 times faster than the previous models at identifying kilns, while crucially maintaining or increasing the accuracy of the model. Testing across regions in Pakistan, India and Afghanistan demonstrated the scalability of this work, and potential application to identifying other objects, like industrial units and oil tanks. By allowing kilns to be identified more quickly and efficiently, this work supports planetary health efforts globally, and the UN’s Sustainable Development Goals on air pollution and forced labour.

To hear more about the impact of this research, read the coverage from the University of Oxford News Team. You can also read a summary of the research in the conference paper prepared in 2023.