Adaptive Stacked Denoising Autoencoder for Work Mode Identification of Airborne Active Phased Array Radar

This paper proposes a new method to recognize airborne phased array radar (AESA) under different modes, based on multi-level modeling combined with Adaptive Stacked Denoising Autoencoder. In order to analyze the change law of pulses intercepted by intelligence, multi-level modeling is proposed to model the pulses at pulse level, pulse group level and work mode level. Then adaptive stacked denoising auto-encoder is trained to extract amplitude characteristics at the work mode level. Finally Softmax classification is added to the top of deep network to realize work mode recognition of airborne phased array radar. Qualitative experiments show that compared with the original algorithm based on knowledge base, the new method is able to extract essential characteristics of the input, reduce the dependence on prior knowledge, and achieves good performance.