Toward smart ocean monitoring: Real-time detection of marine litter using YOLOv12 in support of pollution mitigation

Marine litter has emerged as a pressing global environmental and public health crisis, posing severe threats to biodiversity, food security, and coastal economies. Effective large-scale monitoring and early detection are critical for mitigating marine pollution, yet current manual and sensor-based approaches are limited by high costs, low efficiency, and insufficient accuracy across diverse marine environments. This study presents a real-time marine litter detection framework based on the latest YOLOv12 algorithm to address these challenges. We developed a multi-class annotated dataset comprising 15 representative marine litter categories using both aerial and underwater imagery. The proposed model integrates attention-enhanced convolutional modules, multi-scale feature fusion, and Distribution Focal Loss to improve detection performance under complex oceanic conditions. Experimental results demonstrate that YOLOv12 achieves an mAP@50 of 0.8354 and mAP@50–95 of 0.7025, with robust performance in the presence of occlusion, reflections, small-object detection, and multi-object coexistence. Visual and quantitative evaluations confirm the model's potential for real-world deployment in autonomous platforms such as UAVs and underwater robots. This work offers a scalable and high-precision solution for marine litter monitoring, providing critical technical support for pollution mitigation, environmental governance, and sustainable ocean management.