Cascade detector for oriented airplane detection in remote sensing images

Although the current target detection field has made remarkable progress, airplane detection in large field optical remote sensing images is still challenging. To this end, we innovatively propose an oriented airplane detector to solve the two difficulties. In response to the slow detection speed of optical remote sensing images, we use an anchor-free structure to avoid complex calculations related to anchor boxes and significantly reduce the use of memory and time in computing. Since the airplane target are distributed with arbitrary orientation, the accuracy of the general detector is reduced. We alleviate this problem by replacing the standard convolution with asymmetric convolution. In addition, we use a cascade structure in prediction, which can further improve accuracy with almost no increase in the calculation. Abundant experiments and detection results show our methods effectiveness. In airplane detection task, our proposed model achieves 88.14 mAP on DOTA-v1.0 and 90.46 on UCAS- AOD.