Learning-Aided Aircraft Detection for High-Resolution SAR Images

The constant false alarm rate (CFAR) detection is one of the most widely utilized detection algorithms in SAR images. However, due to the complexity of the airport scene, using CFAR only cannot achieve satisfying aircraft detection. In this paper, we propose a novel framework of learning-aided aircraft detection by machine learning to overcome the issues. This framework links the classic CFAR with machine learning. While preserving the efficiency of the CFAR, the proposed method introduces machine learning to improve accuracy. Specifically, CFAR and a priori knowledge assist us quickly find the location of potential aircraft targets in the scene, which ensures the fast processing speed of framework. In machine learning, we classify the slices of potential targets and then train a classifier for identifying aircraft. Machine learning algorithms perform well on the classification problem, so the high false alarm caused by the traditional method can be smoothly suppressed. Meanwhile, machine learning algorithms improve the accuracy of the framework. Experimental results verify the significant performance of the proposed aircraft detection framework.