Improve YOLOv5s object detection algorithm for aerial images
Aiming at the problems of complex background,excessive redundant information and undetectable small targets in aerial images,an improved object detection algorithm(GGS-YOLOv5)of YOLOv5s is proposed.Firstly,the GAM attention mechanism is added to the Backbone network to reduce the interference of complex backgrounds,suppress redundant information,focus on detection targets,and enhance the feature extraction ability of the model,and a new structure,SPPFCSPC,is proposed to enhance the receptive field while improving the detection speed and accuracy.Secondly,the GSConv module is introduced in the Neck network to reduce the loss of semantic information and enhance global perception and feature fusion capabilities.Finally,the loss function is replaced with SIoU,and the angle penalty cost is added to effectively reduce the degree of freedom,further improve the convergence speed and detection accuracy of the model.The results of the algorithm ablation in the SeaDroneSee dataset and the comparative experimental results show that the proposed algorithm improves the recall rate by 4.9%and mAP 0.5 by 2.8%compared with the original YOLOv5s.
万方数据
维普
