To solve the problem of low ship identification accuracy in high-resolution remote sensing images,an improved YOLOv7-OBB ship identification method is proposed.The introduction of directional detection frame OBB(Oriented Bounding Box)and KLD loss can effectively solve the problem of missing detection caused by the dense arrangement of ships,slender proportions and arbitrary directions,and retain the target direction information of ships while improving positioning accuracy.The hybrid attention module ACmix is added to the backbone network of the YOlOv 7 basic framework to enhance the sensitivity of the network to small target detection and improve the detection accuracy of small vessels.Adding global attention mechanism(NAMAttention)and Partial convolution(PConv)to the neck can improve the ability of PAN networks to capture key features in complex backgrounds while ensuring model lightweight.Experimental results show that compared with the YOLOv 7 model,our method achieves 88.5%average accuracy,93.0%accuracy,and 84.7%recall on the DOTAships dataset,which is 5%,0.9%,and 3.9%higher than YOLOv7,respectively.It can be proved that compared with the current mainstream algorithms,the method has a significant improvement in detection effect.
NETL
NSTL
万方数据
