Object Detection Algorithm of Drone Image Combining Attention Mechanism and Context

In order to improve the drone image target detection accuracy in complex scenes, this paper combines the attention mechanism and context method on the basis of YOLOv5 to build the drone aerial image target detection algorithm. By fusing the LKA attention mechanism, the model focuses on the target region to be detected to obtain more details of the target, and combines the bidirectional feature fusion network with learnable weight parameters to fuse the context information, better improving the detection performance of small targets. To further improve the performance of the model, the article also proposes the Eagle-IoU loss, which effectively solves the problems of unstable gradients and slow convergence of the original model during early training.