<![CDATA[Dolphin monitoring plays a crucial role in maintaining the balance of marine ecosystems and supporting the ecotourism sector. However, in practice, automated dolphin monitoring still faces significant challenges, particularly when deployed in real-time applications within dynamic underwater environments. Previous research on computer vision-based dolphin detection generally uses models with high computational complexity. This condition has resulted in increased resource requirements and long inference times, making it difficult to apply to underwater device-based monitoring systems with limited computing power. Therefore, it is necessary to develop more efficient detection models and algorithms so that the system can operate reliably under real-world monitoring scenarios in resource-limited environments. Moreover, the adoption of the latest-generation lightweight detection architectures in aquatic scenarios remains limited. To address these challenges, this study proposes the application of YOLOv11-Nano as a lightweight detection architecture designed for low-latency dolphin monitoring on resource-constrained devices. The proposed model is optimized to strike a balance between inference speed and detection accuracy, enabling competitive performance under challenging underwater conditions. Experimental results show that YOLOv11-Nano achieves a computational complexity of 6.4 GFLOPs with 2.59 million parameters, while attaining 65.0% mAP@50, 43.1% mAP@50:95, and an inference speed of 18.34 FPS. These results show that YOLOv11-Nano is capable of delivering stable and efficient performance with relatively low computational requirements and high inference speed, demonstrating strong potential for application in real-time monitoring systems based on devices with limited resources to support automatic dolphin detection as part of marine ecosystem conservation efforts.]]>
