Detection of Illegal Waste Dumping via Satellite Imagery using U-Net and Open-Source Tools Welly Anak Numpang (1*), Noryusdiana Mohamad Yusoff (1), Siti Muazah Md Zin (1), Muhammad Hazrul Haiqal Abdul Wahab (4), Nurul Izza Zainal (1), Siti Nor Afzan Abdul Habib (2) and Kamaruzzaman Wahid (3)
(1) Research Officer, Strategic Application Division, Malaysian Space Agency (MYSA), Malaysia
(2) Research Officer, ICT Development and Geoinformatics Division, MYSA, Malaysia
(3) Director, Strategic Application Division, MYSA, Malaysia
(4) Protege, Strategic Application Division, MYSA, Malaysia
*welly[at]mysa.gov.my
Abstract
Global solid waste generation is projected to increase from 2.24 to 3.88 billion tonnes by 2050, with more than one third currently mismanaged through open dumping or burning, posing severe environmental and public health risks. In Malaysia, nearly 39,000 tonnes of waste are produced daily, and many landfills are already at full capacity, resulting in widespread illegal dumping. Traditional monitoring methods, which rely on manual surveys and visual interpretation of satellite images, are time-consuming, resource-intensive, and highly dependent on the expertise of analysts. Malaysian Space Agency (MYSA) began applying satellite imagery to detect illegal waste dumping sites in 2019, following the Kim-Kim River incident, using SPOT-6/7 (1.5 m) and Pleiades (0.5 m) for visual interpretation. With recent advances in Artificial Intelligence (AI), particularly Machine Learning (ML) and Deep Learning (DL), MYSA has transitioned towards automated detection of illegal waste dumping using Pleiades Neo imagery (0.3 m). In this study, a U-Net architecture with a ResNet34 backbone was implemented in open-source software and trained on 2,213 image chips. The model achieved a final accuracy of 99.96% with no signs of overfitting, and was further validated through field deployment in Penang, where verification accuracy reached 96.86%. The model successfully improved overall accuracy by eliminating misclassified outliers, particularly those arising from cemetery areas. By integrating advanced DL and remote sensing technologies, manpower requirements were reduced by up to 99%, leading to a corresponding reduction in labor costs of approximately 93%, significantly enhancing operational efficiency. Looking ahead, MYSA will operationalize the model through the Sistem Pemantauan Potensi Lokasi Pelupusan Sisa (e-Sisa) to support local enforcement and improve waste management. This initiative, aligned with the Malaysia Space Exploration 2030 (MSE2030) Action Plan.
Keywords: deep learning, Pleiaides Neo, remote sensing, illegal waste, u-net
