An Adaptive Complex LISTA Network Exploiting Toeplitz Structure for TomoSAR 3D Reconstruction

In the 3D reconstruction of urban areas using TomoSAR, buildings often exhibit significant sparsity in the elevation direction, typically with only 1 to 3 scattering targets. This inherent sparsity provides a critical prior constraint for enhancing the accuracy of 3D reconstruction result. In this paper, we proposed an Adaptive Complex-valued Learned Iterative Shrinkage and Thresholding Algorithm (LISTA) network with Toeplitz-structured (Ada-CLISTAT). The network significantly enhances estimation accuracy through the incorporation of an adaptive threshold mechanism that effectively constrains the number of targets in the elevation direction. Experiments on simulated and real measured datasets demonstrate that Ada-CLIST AT achieves higher accuracy in estimating the positions and amplitudes of targets in the elevation direction.