摘要
针对遥感图像飞机目标检测问题,探讨了深度卷积神经网络模型Faster R-CNN在遥感图像中对飞机目标检测的应用。针对训练样本不足的问题,构建了Airplane-2018数据集,基于该数据集采用迁移学习的方式对Faster R-CNN模型进行训练,并在测试集上进行验证,在查全率达到95%的情况下,查准率可以达到85%。实验结果表明,Faster R-CNN模型在采用迁移学习方法训练后,在遥感图像飞机目标检测问题上具有可行性。
In view of the aircraft target detection in remote sensing images,the application of Faster R-CNN model used in aircraft target detection of remote sensing images is discussed.In order to solve the insufficiency of training samples,a novel Airplane-2018 dataset is built.The Faster R-CNN model is trained by using transfer learning method based on the proposed dataset,and the validation is carried out on the test set.The precision rate can reach 85% when the recall rate approaches 95%.The experiment results show that the Faster R-CNN model is feasible for aircraft target detection in remote sensing images after training with transfer learning method.
作者
常鹏飞
段云龙
CHANG Pengfei;DUAN Yunlong(The 27th Research Institute of CETC,Zhengzhou 450047,China)
出处
《无线电工程》
2019年第10期925-929,共5页
Radio Engineering
