SM-LightNet: A Lightweight Ship Detection Algorithm Based on Improved YOLOv8

This article presents a streamlined ship detection algorithm SM-LightNet based on YOLOv8 improvement, which is suitable for environments such as ports with dense ship traffic and complex maritime navigation. Firstly, we introduce a Slim-neck structure that includes a lightweight convolution called GSConv and VoVGSCSP structure. It replaces the traditional convolution in the Neck part of YOLOv8 with GSConv, achieving real-time detection of edge devices and effectively reducing the complexity of computation and parameters. Secondly, replacing the C2f structure in Neck with VoVGSCSP enables the model to have a more concise network structure, which can effectively fuse feature information and enhance the precision and stability of object detection. Finally, the loss function MPDIoU was introduced, which calculates IoU by minimizing the point distance between the estimated boundary box and the actual boundary box, simplifying the calculation process and improving detection accuracy. The outcomes of the experiments indicate that the accuracy of the lightweight algorithm reaches 98.2%, and the GFLOPs of the algorithm are 7.4, which is 9.5% lower than the performance of the original model. The model parameter size is 2.80M, which is 7.5<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> lower than the original model.