An easy and fast method for landfill identification by image-based deep learning

Landfills are fundamental urban infrastructures, yet improper operation causes negative impacts on the environment and public health. The accurate geographic information of landfills is often lacking, limiting effective monitoring and management. We develop a methodology that leverages remote sensing and deep learning to efficiently identify landfill locations from Google Maps, which includes: (1) creating a multi-resolution image database of landfill and similar features; (2) introducing a plug-and-play target detection module based on contrastive learning to improve the model's ability to distinguish similar targets and landfills. Experimental results show that using the landfill image dataset with a spatial resolution of 2.15 m can improve detection speed and storage efficiency while ensuring detection accuracy. InternImage-CL achieves the best mAP@.5 of 0.817 with an acceptable training time of 12.75 h at this dataset. This study presents an efficient and scalable method for identifying landfills, providing a methodological basis for digital landfill management and policy development.