TPSSI-Net: Fast and Enhanced Two-Path Iterative Network for 3D SAR Sparse Imaging.

Area of Science:

• Signal Processing
• Electromagnetics
• Computer Vision

Background:

• Compressed Sensing (CS) and 3D Synthetic Aperture Radar (3D SAR) imaging offer potential for reduced sampling rates and enhanced image quality.
• Conventional CS algorithms for 3D SAR face challenges with high computational demands and complex parameter tuning.

Purpose of the Study:

• To develop an efficient and effective framework for 3D SAR sparse imaging.
• To address the limitations of conventional CS-driven algorithms in terms of computational cost and parameter tuning.

Main Methods:

• A two-path iterative framework, TPSSI-Net, is proposed, mapping the Approximate Message Passing (AMP) algorithm to a deep neural network.
• TPSSI-Net incorporates a two-path Convolutional Neural Network (CNN) for nonlinear sparse representation and features learnable coefficients for modified Onsager terms.
• The network is trained end-to-end using a channel-wise loss function that enforces symmetry constraints and measurement fidelity.

Main Results:

• TPSSI-Net demonstrates effectiveness and high efficiency in 3D SAR sparse imaging.
• The proposed method successfully reduces computational costs compared to traditional CS algorithms.
• Experimental results from simulations and real-measured data validate the performance of TPSSI-Net.

Conclusions:

• TPSSI-Net provides a powerful and efficient solution for 3D SAR sparse imaging.
• The deep learning approach overcomes the limitations of conventional CS methods.
• The framework shows significant promise for advancing 3D SAR imaging applications.