Research on maritime ship target detection based on the optimized YOLOv8 model

Accurate detection of maritime ship targets remains a critical challenge in marine surveillance systems. This study proposes an improved YOLOv8 model named YOLOv8_optimize to address this challenge. First, we construct a comprehensive dataset comprising over 80,000 high-resolution images covering diverse maritime scenarios and meticulously annotated. Next, the backbone network of YOLOv8 is optimized by replacing the original C2f module with the MBConv module and incorporating depth-wise separable convolution techniques. These modifications significantly reduce computational complexity while maintaining model performance, thereby improving operational efficiency. Furthermore, we refine the loss function of the detection head by adopting the Focal Loss function to mitigate class imbalance issues, enabling the model to prioritize difficult samples and rare categories during the training process, thus significantly enhancing the overall detection performance. Comparative experiments demonstrate that the YOLOv8_optimize model outperforms both YOLOv8n and YOLOv8s in ship detection tasks, offering an efficient and robust solution for maritime applications with substantial practical value.