融合双分支网络与视觉引导注意力的遥感小目标检测OA

To address faint features and complex backgrounds for small objects in remote-sensing imagery,a dual-branch network(visual-guided dual-branch attention YOLO,VGDA-YOLO)that fuses visible light and infrared was proposed.Based on YOLO11,A dual-branch backbone network combining the advantages of YOLO and Transformer was designed by this model.It processes visible light and infrared data in parallel,enabling efficient capture of local details of targets and global contextual dependencies.To address the difficulty in extracting structural features of small targets,a biomimetic directional perception module was presented.By simulating the directional selectivity of the primary visual cortex,it enhances the ability to extract structural features of small targets.To solve the problem of background interference,a visual guided attention fusion module was proposed.Imitating the"fovea-periphery"coordination mechanism of human vision,shallow detail features as guidance to perform selective weighted fusion of deep semantic features was used,effectively suppressing background noise and achieving accurate feature fusion.Experimental results on the public VEDAI(vehicle detection in aerial imagery)dataset demonstrate that VGDA-YOLO achieves a mean average precision of 0.750,outperforming multiple benchmarks and advanced detection models.