Star-YOLO: a lightweight and detail-aware network for infrared ship detection

Infrared ship detection plays a vital role in maritime security and autonomous navigation, particularly under low-light conditions or in complex sea environments. However, existing object detection models face challenges when applied to infrared imagery, including high computational cost, weak performance in small-object detection, and the loss of fine-grained details. To address these issues, this paper proposes a lightweight infrared ship detection framework based on YOLOv11, termed Star-YOLO, which incorporates three structural improvements to enhance detection accuracy while reducing model complexity. First, a lightweight backbone named StarNet replaces the original YOLOv11 backbone, reducing parameters and computational overhead while maintaining strong feature extraction capability. Second, to better detect small targets commonly missed in infrared imagery, a lightweight bottleneck module integrating ghost modules and dynamic convolution replaces the original bottleneck in the C3k2 module, enhancing the network's sensitivity to small-object features. Finally, enhanced linear deformable convolution substitutes standard convolutional layers, further compressing model size and improving the representation of fine-grained details in infrared scenes. Experimental results on both the infrared ship detection dataset and NUDT-SIRST-Sea demonstrate that the proposed model significantly reduces parameter size and computational cost while maintaining high detection accuracy, making it well-suited for edge deployment and practical maritime applications.