Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{178780,
        author = {Munnanuri Akshitha and Gajula Yashwanth and Battu Srinath Reddy and Rishikesh Reddy Boyapally and Mohammed Afzal and Dr. M . Ramesh},
        title = {Efficient Deep Learning-Based Marine Plastic Waste Detection with YOLOv12},
        journal = {International Journal of Innovative Research in Technology},
        year = {2025},
        volume = {11},
        number = {12},
        pages = {6465-6472},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=178780},
        abstract = {Marine plastic pollution threatens marine ecosystems, necessitating scalable and accurate detection techniques for effective mitigation. A novel YOLOv12l-based approach to detect and quantify plastic waste in underwater environments, solving the issues of submerged plastics in complex underwater imagery, is put forward in this study. With a YOLO-typed dataset gathered using Roboflow, consisting of 4,398 training images, 386 validation images, and 205 test images distributed across four classes—trash, plastic, metal, and glass—the model was trained for 50 epochs on a dynamic learning rate, achieving a mean average precision (mAP) of 90.2%. The dataset has underwater augmentations such as blur and noise to simulate real marine environments, hence the robust model. The system performs better with 93.5% plastic detection correctness, although metal detection is 88.7% due to reflective issues, showing stable feature extraction in the presence of diverse underwater scenarios like aquatic life. With edge deployment tuned on a Raspberry Pi 4, the system offers 10 frames-per-second real-time inference, supporting continuous monitoring with Camera Module 2 support for GPS for precise debris mapping. Validation loss collapsed to 0.52, and test performance suggests outstanding generalization, outperforming YOLOv11m (87.6% mAP). Limitations are the amount of training images (4,989), which may limit generalization to uncommon classes of debris. This research strongly enhances autonomous ocean cleanup through a light-weight, high-precision system, with significant potential for environmental monitoring, policy support, and sustainable marine management.},
        keywords = {Ocean Plastic Detection, YOLOv12l, Underwater Imagery, Multi-Class Detection, Environmental Monitoring},
        month = {May},
        }