Extraction and Analysis of RFI Signatures via Deep Convolutional RPCA

Radio Frequency Interference (RFI) poses a significant threat to microwave remote sensing instruments like synthetic aperture radar (SAR), which causes information loss, image degradation and reduces measurement accuracy. In this paper, considering the temporal-spatial correlation of target response, and the random sparsity property for time-varying interference, we propose a novel approach for mitigating RFI signals in SAR raw data utilizing the joint low-rank and sparse property. Instead of applying the iterative optimization process with uncertain computation burden, the proposed Deep Convolutional RPCA approximates the iterative process with a stacked recurrent neural network. It employs the supervised deep learning to speed up the efficiency and adjusts the hyperparameters adaptively. The experimental results show that the validity of the proposed method.