GAENet: A Global Attention-Aware Enhanced Network for Remote Sensing Target Detection

Due to the large-scale variation of targets and complex background in optical remote sensing images, existing remote sensing target detection algorithms face significant challenges. To address these issues, a global attention-aware enhanced network (GAENet) for remote sensing target detection is proposed. Firstly, the convolution and attention fusion module (CAFM) is introduced in the backbone network, which effectively combines the convolution and attention mechanism modules in feature modeling, enhancing the long-distance dependency between features and the feature expression ability of local regions. Secondly, the dynamic detection head (Dyhead) is introduced, which adaptively adjusts the attention distribution to improve the model’s perception ability for targets of different scales. Experimental results on the DIOR dataset show that the proposed method can significantly improve the detection accuracy of multi-scale targets in complex scenes, achieving state-of-the-art performance. The source code is available at https://github.com/object-detection/GAENet.