Key points are not available for this paper at this time.
Aircraft detection in synthetic aperture radar (SAR) images is a tough problem in remote sensing image interpretation, due to different scattering characteristics and scales of aircrafts as well as the influences of airport metal objects. Although deep convolutional neural networks perform well in object detection on optical images, the application of deep learning for SAR image aircraft detection has not been adequately studied. In this article, a cascaded three-look network is proposed to locate aircrafts in SAR images. The detection networks for the airport and aircrafts are modified separately in region proposal, which combines the superiorities of Faster R-CNN in object detection and residual units in feature extraction. The network contains three looks in noticing airport, detecting aircraft and extracting airfield runway. Three looks are designed not only to reduce the detection range and false alarms but also to elevate the detection precision. Experiments conducted on test data demonstrate that the proposed method can achieve 0.67 in F1 score, which outperforms the compared method.
Zhang et al. (Wed,) studied this question.