针对船舶号灯目标检测中参数计算量大、背景光线复杂及号灯种类繁多等问题，对YOLOv8n做出了改进，以满足实时准确识别船舶号灯的要求。首先使用VanillaNet网络作为主干特征提取网络，以降低模型的计算成本，满足实时检测要求；其次引入颜色注意力模块以确定和增强号灯的颜色特征，以提高在复杂光线背景下号灯颜色的识别能力；再次，针对号灯的大小、频率、各视图空间排列等特征，构建MoE-layer模块代替C2f模块，以提高船舶号灯识别的准确性；最后通过Focal Loss调节对难易分类样本的关注来解决类别不平衡问题，提高模型对号灯目标检测能力。实验结果表明，相较于原基线YOLOv8n模型，改进后模型在参数量和计算量分别降低37.7%和52.8%的情况下，精度和mAP@0.5分别提升3.3%和2.2%，达到98.3%和98.7%，可以满足实时识别船舶号灯的要求。

This study addresses the challenges associated with high computational parameter volume, intricate complex background lighting, and the diversity of the ship's lights in the context of target detection. improvements have been made to YOLOv8n to meet the requirements for real-time and accurate identification of ship’s lights. Initially, we employ the VanillaNet as the core feature extraction network, which significantly diminishes the model's computational overhead and ensures compliance with real-time detection demands. Subsequently, we integrate a color attention module designed to discern and amplify the distinctive color attributes of the navigation lights, thereby bolstering their recognition even amidst complex lighting conditions. Furthermore, to accommodate the unique characteristics of the ship's lights, including their size, frequency, and spatial distribution across different perspectives, introduce a Mixture of Experts (MoE)-layer module as a substitute for the conventional C2f module, further refining the identification accuracy. Finally, the Focal Loss function is adopted to adjust the focus on easy-to-hard classified samples, addressing the class imbalance issue and improving the model's ability to detect ship’s lights. Experimental results demonstrate that, compared to the original baseline model YOLOv8n, the improved model reduces parameters and computation by 37.7% and 52.8%, respectively, while increasing accuracy and mAP@0.5 by 3.3% and 2.2% to 98.3% and 98.7%, respectively, and therefore the improved YOLOv8n meets the requirements for real-time identification of ship’s lights.