Remote Sensing for Risk Management: Solid Waste Detection Using YOLOv10

Abstract

The effective detection of obstructions in urban water channels is critical for mitigating social and environmental impacts, particularly those related to flooding, and for strengthening environmental management. Traditional methods employed by local authorities for identifying solid waste are often expensive and pose health risks to field personnel. This study presents a system for detecting obstructive solid waste in urban waterways using the YOLOv10 deep learning model. Leveraging drone imagery collected over a coastal city's drainage channel, the approach generates reliable data to support early detection of flood-related hazards. A dataset of 1230 augmented images was used to fine-Tune a pre-Trained YOLOv10x model over 150 epochs. Preliminary results show the model's ability to effectively identify garbage, debris, and vegetation, offering an innovative tool for urban flood risk mitigation. The findings highlight the potential of deep learning to enhance environmental monitoring and support the planning of more resilient urban infrastructure.