Automated Detection and Mapping of Waste in Different Environments and Seasons Using UAV Imagery and AI

Plastic pollution caused by illegal dumping in peri-urban and rural environments poses significant environmental challenges and remains difficult to monitor using traditional field surveys or medium-resolution satellite imagery. This study presents an automated framework for detecting and mapping waste across different environments and seasons using UAV full-motion video, very high-resolution imagery and artificial intelligence. RGB and thermal videos and photographs acquired with a DJI Mavic 2 Enterprise Dual UAV were processed to generate datasets with 1–1.5 cm spatial resolution. Training samples representing plastic litter, construction materials and rubber were manually digitized and used to train state-of-the-art object detection models implemented in Python. Two complementary paradigms were evaluated: YOLO, optimized for real-time detection speed from video streams, and DETR, designed for higher detection accuracy and robustness in complex scenes. The approach was applied in multiple Romanian regions (Giurgiu, Călărași, Constanța) across diverse landscapes, including pastures, sandy areas, forest edges and recreational sites. Over 3000 waste objects were automatically identified, with an overall detection accuracy of 0.91 and confidence levels between 80% and 99%. Spatial analysis revealed that waste accumulation is strongly associated with secondary roads, access routes and intersections, indicating vehicle-based dumping patterns consistent across environments. The results demonstrate that integrating UAV video, high-resolution imagery and advanced AI enables efficient, scalable and reproducible monitoring of illegal waste. The framework supports rapid identification of pollution hotspots and can inform targeted clean-up actions and environmental management strategies, representing a promising Earth Observation solution for high-resolution terrestrial pollution monitoring.