Optical Remote Sensing Ship Detection Based on Improved YOLOv8

In light of the characteristics of ship targets with large-scale changes and harrowing feature extraction in the optical remote sensing images, we propose an improved ship detection algorithm for optical remote sensing based on YOLOv8. Firstly, from feature extraction, we propose a multi-scale grouped feature extraction module called C2f-MSG, which mitigates the computational burden by eliminating redundant feature maps and prioritizing the extraction of multi-scale ship features. Secondly, from regression loss, we design the IoU loss called Inner-MPDIoU, which improves the overlapping bounding box regression by simplifying the similarity comparison among two bounding boxes and generates gains for ship samples with high IoU by controlling the size of the auxiliary bounding box. Finally, the experimental results show that the detection accuracy of the improved model has increased. Additionally, the model showcases a reduction in parameters and computing effort, exemplifying its superiority in terms of accuracy and complexity.