Automatic Target Recognition on High Resolution SAR Images with Deep Learning Domain Adaptation

Synthetic Aperture Radar (SAR) images provide a good means of analyzing large areas despite fog, clouds, etc. High resolution SAR images enable the detection of a wide variety of relevant targets to detect potential threats and assess the situation. However, the vast amount of information and the peculiarities of SAR imagery require a huge amount of trained experts and/or time to comb through the data. To reduce this massive time and cost factor automatic target recognition is required to detect and classify potential targets. With the recent advances in the field of deep learning based domain adaptation, the automatic detection can now be achieved on an unprecedented level. We demonstrate, how simulated data can be used to almost eliminate the need for the rare and very cost intensive real target data. In particular, we show that even without any real training data, a very good classifier can be trained.