Improved YOLO Model for Object Detection of Underwater Trash

Accurate detection of underwater trash is crucial for effective marine conservation efforts. In this paper, we present an improved You Only Look Once (YOLO) model specifically tailored for object detection of underwater waste. In the network structure, the imbalance of feature information due to the splicing operation on the front side of the detection head is solved by integrating the ConvNeXt block into the YOLO model; in the backpropagation stage, the localization error of the prediction box is reduced by adopting EIoU Loss as the model loss function. our model demonstrates significant improvements in both detection accuracy and robustness compared to existing approaches. Experiments conducted on underwater trash images showcase the effectiveness of our model, achieving state-of-the-art performance.