A Lightweight Object Detection Model for Complex Marine Conditions Based on Improved RT-DETR

Object detection in complex maritime environments is particularly challenging due to factors such as dynamic backgrounds, varying object appearances, and limited computational resources available for deployment. This paper focuses on developing a lightweight and real-time object detection solution for such conditions. To address these issues, we propose a novel framework named MRL-DETR (Marine Real-time Lightweight DEtection TRansformer), built upon the RT-DETR architecture. The proposed model introduces two key innovations: (1) the integration of MobileNetV3-small as a lightweight backbone to effectively reduce parameter count and computational cost; and (2) the incorporation of a Dilated Reparameter Block (DRBC3) into the detection head to enhance semantic feature representation and enlarge the receptive field, while maintaining high inference efficiency. The objective of this work is to achieve a balanced trade-off between detection accuracy, speed, and resource consumption, suitable for deployment in real-world maritime applications. Experimental evaluations demonstrate that MRL-DETR reduces model parameters by 60.6% and FLOPs by 73.4% compared to the baseline RT-DETR-R18 model, while maintaining competitive accuracy and significantly improving detection speed. These results validate the effective-ness and deployment potential of MRL-DETR for maritime object detection tasks.