Illegal Waste Dump Detection Using a Deep Learning Approach: Performance Evaluation of YOLO Architectures

Illegal waste dumps present significant environmental and public health risks, especially in economically disadvantaged areas that lack sufficient waste management systems. This study examines the efficacy of five state-of-the-art YOLO object detection models—YOLOv8, YOLOv9, YOLOv10, YOLOv11, and YOLOv12—in identifying unlawful waste dumps in diverse outdoor scenarios and utilizing a meticulously collected and enhanced library of annotated garbage images from several environments (urban, coastal, and roadside). The models' performance was evaluated using conventional object detection measures, including Precision, Recall, F1-Score, mAP@50, and mAP@50—95. YOLOv9 attained the greatest F1-Score (87%), YOLOv11 excelled in mAP@50 (88%), and YOLOv10 had superior performance in mAP@50–95 (65%), although YOLOv8 and YOLOv12 displayed consistent robustness under diverse situations. Based on qualitative analysis, context-dependent strengths and limitations are associated with the architectural design of each model. The results highlight the compromises among reliability, consistency, and inference efficiency, offering pragmatic guidance for model selection in real-time environmental monitoring. This paper highlights the significance of deep learning-based computer vision for automated waste dump detection and advocates for future integration with Unmanned Aerial Vehicles or surveillance systems in waste management efforts.