<p>Under the complex maritime environments, infrared ship detection plays a crucial role in maritime rescue and traffic control. Unlike traditional visible-light imaging, infrared imaging provides reliable target detection even under low-light conditions and adverse weather. However, it is difficult for existing detection methods, especially deep learning methods, to achieve high detection accuracy with low model complexity. To address this issue, the fast precision real-time detection transformer (FP-RTDETR) network is proposed for infrared ship detection. Specifically, two novel modules are designed in FP-RTDETR: the focused feature pyramid network (FFPN) and the multi-scale parallel aggregation splicing (MPAS). The FFPN is designed to effectively capture long-range dependencies, which is crucial for enhancing the detection of small targets. The MPAS module extracts multi-scale infrared information, thereby improving the network’s adaptability to varying target sizes. In addition, to enhance inference efficiency, the Star_block module is incorporated to simplify the network architecture. Moreover, a new module, AIFI-RepBN, is designed by integrating reparameterized batch normalization (RepBN) into the Attention-based intra-scale feature interaction (AIFI) module, which can significantly improve inference speed. We also introduce inner-EIoU into real-time detection transformer (RT-DETR) network to boost detection accuracy, which combines inner-intersection over union (Inner-IoU) and efficient intersection over union (EIoU) with a scaling factor for auxiliary bounding boxes. Experimental results show that FP-RTDETR improves mAP50 by 1.5%, mAP50:95 by 1.5%, and FPS by 21.2%, while reducing GFLOPs by 13.5%. On the HIT-UAV dataset, mAP50 increases by 0.9%, and on the CTIR dataset, it improves by 1.8%.</p>