RT-DETR-FCES: a lightweight ship detection algorithm from remote sensing perspective

Ship detection plays a crucial role in maritime traffic safety, emergency rescue, and intelligent port management. However, existing methods still face challenges in balancing detection accuracy, real-time performance, and lightweight deployment in complex maritime environments, especially for small ship targets on resource-constrained platforms. To address these issues, this paper proposes RT-DETR-FCES, a lightweight real-time ship detection framework for optical remote sensing images. Specifically, FasterNet is adopted as the backbone network to reduce computational cost and model parameters, while Cascaded Group Attention (CGA) is introduced to enhance multi-scale feature fusion. In addition, an EUCB-SC module is designed to improve feature representation capability with efficient inference. Experimental results show that RT-DETR-FCES achieves 87.5% \(mAP_{50}\) on the MASATI-v2 dataset and 49.1% \(mAP_{50}\) on the VAI dataset. Compared with the baseline RT-DETR, the proposed method reduces the number of parameters and FLOPs by 45.5% and 48.0%, respectively, while achieving inference speeds of 87.7 FPS and 85.1 FPS. Overall, the proposed framework achieves a favorable balance between detection accuracy and lightweight efficiency, demonstrating its effectiveness for real-time maritime ship monitoring applications.