Remote sensing image detection method based on context-aware mechanism and transformer architecture

Remote sensing image object detection faces numerous challenges, such as complex backgrounds, multi-scale target recognition, and high-resolution image processing. This paper proposes an RT-DETR model based on an improved transformer architecture, aiming to enhance detection accuracy and robustness. First, a MogaC2f module is introduced, which effectively improves the representational capability and computational efficiency of feature extraction. Second, a ContextGuidedBlock module is designed, which integrates context-aware mechanisms with downsampling operations to enhance the model’s sensitivity to object boundaries and local details, thereby improving small object detection performance. Finally, the proposed AIFIMSMHSA module leverages adaptive feature interaction and a multi-scale multi-head self-attention mechanism to strengthen spatial perception in high-level features, achieving precise localization across different target scales. Experimental results demonstrate that the proposed model outperforms the existing mainstream methods across several remote sensing object detection datasets, with especially notable improvements in small object detection and complex scene understanding. In addition, the model exhibits superior computational efficiency and fewer parameters than the current state-of-the-art models. Experiments conducted on representative datasets such as NWPU-VHR10, remote object sensing dataset, and small infrared moving detection further validate the model’s strong cross-scene generalization and robustness. Visualization analyses also indicate better boundary fitting and confidence distribution in real-world complex remote sensing scenes, underscoring its application potential in high-resolution remote sensing image object detection tasks.