YOLOv8-Aerial:面向无人机航拍图像的目标检测优化方法OA

Object detection for aerial images is crucial for smart cities and disaster monitoring but faces challenges like complex backgrounds,scale variations among objects,and densely distributed small objects.To address these challenges,a YOLOv8-Aerial model is proposed that systematically optimizes three key components of YOLOv8s:the feature extraction network,feature fusion network,and detection head.First,a C2f_MaskUnitAtt module incorporating a masked attention mechanism is designed to enhance the model's feature extraction capability in complex backgrounds.Next,for feature fusion,a bidirectional adjacent layer feature fusion method combined with a super-resolution enhancement strategy is proposed,to fully exploit multi-scale features and mitigate the feature attenuation problem of small objects in deep networks.For the final detection,a small object-oriented detection head is additionally introduced and a local feature enhancement module with dilated convolution is designed to improve detection accuracy for small objects.Experimental results demonstrate our model's superiority over existing models.