We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Tunisian Mediterranean Coastal Marine Litter and trash Dataset: Drone-Based Real-World Images for Detection, Classification and Segmentation
Summary
Researchers used drones to photograph thousands of pieces of trash, plastic, glass, metal, and more, scattered along Tunisian Mediterranean beaches, then built a labeled dataset to train AI systems to automatically spot and sort marine litter. This matters because plastic waste in our oceans breaks down into microplastics that enter seafood and drinking water, and better AI-powered detection tools could help communities track pollution faster and clean up coastlines before that trash fragments and works its way into the food we eat.
Dataset Description The Tunisian Mediterranean Coastal Marine Litter and trash Dataset (TUN-MarineLitter) is a real-world image dataset developed from drone-based aerial surveys conducted along selected coastal areas of the Tunisian Mediterranean Sea. The dataset was created to support research and development in marine litter detection, classification, and instance segmentation using computer vision and deep learning. The database contains high-resolution aerial images acquired in real coastal environments and includes seven major marine litter categories: Cardboard, Fabric, Glass, Metal, Other, Plastic, and Wood. Objects are annotated using polygon-based labels in normalized YOLO segmentation format, enabling direct use with modern YOLO-based detection and segmentation architectures. This version of the dataset is divided into 80% training, 15% validation, and 5% test sets. Data augmentation is applied exclusively to the training set using controlled ROI rotation (−15° to +15°), light exposure variation (−13% to +13%), and 10% salt-and-pepper noise. The validation and test sets remain unaugmented to ensure reliable and unbiased model evaluation. The images were collected from selected Tunisian Mediterranean coastal study areas in Mahdia, under real-world environmental conditions and varying illumination, viewpoints, backgrounds, and object scales. The selected coastal areas encompassed diverse environmental backgrounds, including sandy beaches, rocky shorelines, Posidonia-covered areas, and hybrid environments combining sandy, rocky, and Posidonia habitats, thereby ensuring a representative range of coastal conditions for robust marine waste detection and classification. The dataset is intended for applications including UAV-based environmental monitoring, marine pollution assessment, autonomous coastal surveillance, object detection, image segmentation, and artificial intelligence for marine environmental protection. The dataset was developed under the scientific leadership of Full Prof. Dr. Ing. Hassene Seddik, Project Leader at the RIFTSI Laboratory (Robotique Intelligente, Fiabilité, Traitement de Signal et de l'Image) (Intelligent Robotics, Reliability, Signal and Image Processing), École Nationale Supérieure d’Ingénieurs de Tunis (ENSIT) (Higher National Engineering School of Tunis) (https://ensit.rnu.tn/) , with the contributions of Mr. Riadh Salah Chaouachi, PhD Researcher; Mr. Moataz Harrazi, PhD Researcher; Ms. Anas Aouadi, PhD Researcher; Mr. Rabie Ouerghie, PhD Researcher. The project also benefited from the scientific collaboration of the associate professor. Noureddine Zaaboub from the National Institute of Marine Sciences and Technologies (INSTM), as well as researchers and collaborators from ENSIT, the RIFTSI Laboratory, and ENICar. Keywords Marine litter; Marine debris; Plastic pollution; Coastal monitoring; Mediterranean Sea; Tunisia; Drone imagery; UAV; Aerial imagery; Computer vision; Deep learning; Object detection; Instance segmentation; YOLO; Environmental monitoring; Marine pollution; TUN-MarineLitter.