1-Bit Radar Imaging Based on Adversarial Samples

Deep unfolding BIHT network with threshold for SAR imaging

“…An intuitive idea of deep unfolding is to use the powerful fitting ability of neural networks to achieve denoising. At the same time, we observed that whether it is the time-varying threshold [5] , adversarial samples [6] or adaptively threshold [12] , they all introduce a threshold but use different processing strategies. Therefore, when building the proposed network from BIHT, a threshold parameter is also introduced.…”

“…The neural network has excellent nonlinear representation ability and has a good performance of denoising, which can be used to solve the problem of poor reconstruction performance of the BIHT algorithm under low SNR. In addition, further considering the same thoughts of time-varying threshold algorithm [5] , adversarial samples algorithm [6] and adaptively threshold algorithm [12] to enhance performance at low SNR, we also introduce a threshold parameter for adjustment to characterize the noise. The iteration expression after introducing the threshold parameter ℎ is (10).…”

“…Demir et al [5] started from the perspective of amplitude information loss in the one-bit quantization process, and introduces the time-varying threshold, thus more amplitude information can be used during the reconstruction process to get good performance at the low SNR. Han et al adopted adversarial adjustment of the threshold parameter during the algorithm's iteration process to reduce the number of false targets under low SNR [6] .…”

Sparse Logistic Regression-Based One-Bit SAR Imaging

“…The binary iterative hard threshold (BIHT) method addresses quantization errors by enforcing consistent reconstruction, which has the advantage of easy implementation and has been demonstrated to be successful and efficient in enhancing the performance of one-bit signal recovery [31], [34], [35]. Building upon this foundation, several researchers have proposed one-bit SAR imaging approaches based on the BIHT method framework [10], [36], [37]. In [36], an adaptive BIHT (A-BIHT) method is proposed, which introduces an adaptive quantization level parameter scheme and iteratively updates the imaging results and quantization level parameters.…”

“…Based on this insight, an enhanced BIHT (E-BIHT) method is proposed, resulting in an image with a higher target clutter ratio. Motivated by the advantages of adversarial samples, [37] introduces the adversarial sample-based BIHT (AS-BIHT) method for onebit radar CS imaging. AS-BIHT formulates the one-bit radar imaging problem by constructing a quantized level parameter within the imaging process and defining the problem as a onebit radar imaging problem with a quantized level parameter as a parametric model.…”