Abstract:

Interrupted sampling repeater jamming (ISRJ) takes advantage of the matched filtering characteristic of synthetic aperture radar (SAR) and causes the fake targets with the interval distribution characteristic in SAR image. These fake targets are a kind of deception for target detection. So, it is quite significant to detect and restrain the ISRJ. However, the corresponding researches mainly focus on the signal domain at the present stage. Therefore, this work researches on the ISRJ detection in the image domain. Firstly, by combining the measured data with the simulated jamming, ISRJ samples are produced and established based on different measured scenes and simulated parameters. Secondly for the interval distribution characteristic of the fake targets, the two-stage and one-stage models, which are representative in the field of deep learning detection are adopted. Thirdly, ISRJ samples from one single scene are used for the models training, and the samples from the other scenes are tested by the well-trained models. Finally, the ISRJ detection results are obtained. Based on MiniSAR data, the experiment verifies that for the ISRJ samples with different kinds, scenes, and parameters, the adopted deep learning models are able to achieve the averaged total detection accuracy of 95.75%. They are with the good generalization abilities. Moreover, for the sample with the size of 501 pixels×501 pixels, the least time cost of these models is 0.035 s.

Key words: synthetic aperture radar (SAR), interrupted sampling repeater jamming (ISRJ), jamming detection, deep learning models