A Refined Deep Learning Method Using Yolo for Marine Debris Identification

Environmental ecosystems are seriously threatened by marine trash, hence efficient techniques for locating and detecting it are required. By putting forth a novel strategy that combines the instance detection capabilities of YOLOv11 with different attention mechanisms to improve efficiency and expand application, this study overcomes the constraints currently present in the literature. While YOLOv11 and Coordinate Attention function well in a variety of contexts, the bottleneck transformer is useful in locating debris regions that are missed via manual annotation, despite being somewhat less consistent overall. Furthermore, the bottleneck transformer exhibits improved accuracy in identifying larger debris fragments, suggesting its possible use in particular applications. The adaptability and effectiveness of attention-enhanced YOLOv11 models for maritime trash identification are highlighted in this work, proving their capacity to handle a range of environmental monitoring difficulties. The findings also highlight how crucial it is to match detection models with certain operational needs because each model has special qualities that can be useful in particular situations. Keywords: YOLOv11, Marine Debris, Coordinate Attention, Bottleneck Transformer, Object Detection, Deep Learning, Environmental Monitoring, Feature Extraction, Instance Segmentation, Drone Surveillance