MDD-YOLO: Marine Debris Detection Algorithm Based on Improved YOLOv11n

Marine debris detection is a challenging task of great significance to marine ecological protection. To address the poor performance of existing detection methods, caused by insufficient lighting and low-resolution underwater images, we propose a new target detection algorithm named Marine Debris Detection-YOLO(MDD-YOLO), based on the YOLOv11n model. First, we introduce Efficient Channel Attention(ECA) module to enhance focus on key features. Then, the C3k2 module with Swin Transformer (C3k2STR) is incorporated to improve the model's ability to capture global features and rich contextual information. Moreover, we integrate SPD-Conv into Efficient Up-Convolution Block (EUCB), replacing the standard up-sampling operation to address the challenge of detecting small objects. Finally, the SIoU loss function is implemented to further optimize the positioning accuracy of the bounding box by incorporating angle penalty cost, dynamic distance cost, and shape consistency constraint. Experimental results show that the proposed MDD-YOLO algorithm significantly improves marine debris detection. On the TrashCan 1.0 for YOLO dataset, it achieves improvements of 3.2% in Precision, 4.7% in mAP@0.5 and 4.3% in mAP@0.5:0.95.