SAR Image Active Jamming Type Recognition Based on Deep CNN Model
Siwei CHEN,
Xingchao CUI,
Mingdian LI,
Chensong TAO,
Haoliang LI
Affiliations
Siwei CHEN
The State Key Laboratory of Complex Electromagnetic Environment Effects on Electronics and Information System, College of Electronic Science, National University of Defense Technology, Changsha 410073, China
Xingchao CUI
The State Key Laboratory of Complex Electromagnetic Environment Effects on Electronics and Information System, College of Electronic Science, National University of Defense Technology, Changsha 410073, China
Mingdian LI
The State Key Laboratory of Complex Electromagnetic Environment Effects on Electronics and Information System, College of Electronic Science, National University of Defense Technology, Changsha 410073, China
Chensong TAO
The State Key Laboratory of Complex Electromagnetic Environment Effects on Electronics and Information System, College of Electronic Science, National University of Defense Technology, Changsha 410073, China
Haoliang LI
The State Key Laboratory of Complex Electromagnetic Environment Effects on Electronics and Information System, College of Electronic Science, National University of Defense Technology, Changsha 410073, China
Synthetic Aperture Radar (SAR) can acquire high-resolution radar images of region of interest under all-day and all-weather conditions, a capability that has been successfully applied in many fields. In the environment of military confrontation games, complex electromagnetic jamming severely impacts SAR image interpretation and intelligence generation. Scholars have proposed numerous SAR anti-jamming approaches to date. However, the recognition of SAR image jamming types, which is the prerequisite of anti-jamming, has rarely been reported. This work focuses on active jamming type recognition in SAR images. First, five typical active jamming modes are selected and further subdivided into nine jamming types based on various jamming parameters, which serve as the objects of jamming recognition. The typical active jamming datasets are then constructed based on the stacking of simulated jamming signal echoes and real-measured MiniSAR data in the echo domain and SAR imaging processing. Based on the jamming datasets, an attention-combining deep Convolutional Neural Network (CNN) model has been proposed. Thereafter, comparative experiments are performed. Experiments show that, compared with traditional deep CNN models, the proposed method achieves more accurate recognition and more stable performance across various scenes and jamming parameter configurations.
