Object detection poses significant challenges for small marine debris due to complex beach environments. The main difficulties are homogeneous and low-contrast backgrounds, color similarities between targets and surroundings, and small objects which are often occluded. To address these challenges, this paper proposes a Small Object Enhanced Architecture (SEA) to the RT-DETR hybrid encoder to improve small object detection performance in beach environments. The enhanced encoder integrates an SPDConv layer into the P2 feature layer to capture spatial information and introduces a CSPOmniKernel module after feature concatenation to build up multi-scale feature representation. Through ablation study and comparison with different models on three distinct datasets focused on marine debris, our RT-DETR-SEA showed a 13.51% increase in mAP@50 compared to the baseline RT-DETR-r18 model and a 9.34% increase in mAP@50 against directly adding a P2 layer. Furthermore, RT-DETR-SEA achieved a similar frames per second (FPS) rate with the baseline model. The source codes are available on GitHub https://github.com/Nathancgy/RT-DETR-SEA