Microwave synthetic aperture interferometric radiometers (SAIRs) are very powerful instruments for high-resolution remote sensing of the atmosphere and the earth surfaces at microwave frequencies. Microwave SAIR imaging reconstruction from interferometric measurements suffers from hardware non-identities, limited prior information, and noise interference, and consequently often requires expert calibration strategies to reduce imaging error and improve the accuracy of the reconstruction. In this article, we propose a new SAIR imaging approach with a deep convolutional neural network (CNN) learning framework to optimize the reconstruction performance. We interpret interferometric measurements of SAIR as a signal encoding representation and SAIR imaging as the corresponding decoding representation. A deep CNN framework with additional fully connected layers is utilized to autonomously learn the decoding representation from interferometric measurement samples and perform SAIR imaging. The supervised learning forward model with hyperparameters makes that the proposed approach could accurately obtain the SAIR imaging representation involving multiple systematic features for real applications. We demonstrate the performance of the proposed imaging approach through extensive numerical experiments. Compared with conventional handcrafted Fourier transform and sparse regularization reconstruction imaging approaches, the proposed imaging approach based on deep learning is superior in terms of image quality, computing efficiency, and noise suppression.

The Compressive Sensing (CS) approach has proven to be useful for Synthetic Aperture Interferometric Radiometer (SAIR) imaging because it provides the same high-resolution capability while using part of interferometric observations compared to traditional methods using the entirety. However, it cannot always obtain the sparsest solution and may yield outliers with the non-adaptive random measurement matrix adopted by current CS models. To solve those problems, this paper proposes a robust reweighted L₁-minimization imaging algorithm, called RRIA, to reconstruct images accurately by combining the sparsity and prior information of SAIR images in near field. RRIA employs iterative reweighted L₁-minimization to enhance the sparsity to reconstruct SAIR images by computing a new weight factor in each iteration according to the previous SAIR images. Prior information estimated by the energy functional of SAIR images is introduced to RRIA as an additional constraint condition to make the algorithm more robust for different complex scenes. Compared to the current basic CS approach, our simulation results indicate that RRIA can achieve better recovery with the same amount of interferometric observations. Experimental results of different scenes demonstrate the validity and robustness of RRIA.

Millimeter-wave interferometric synthetic aperture radiometer (InSAR) can provide high-resolution observations for many applications by using small antennas to achieve very large synthetic aperture. However, reconstruction of a millimeter-wave InSAR image has been proven to be an ill-posed inverse problem that degrades the performance of InSAR imaging. In this paper, a novel millimeter-wave InSAR image reconstruction approach, referred to as InSAR-TVMC, by total variation (TV) regularized matrix completion (MC) in two-dimensional data space, is proposed. Based on the a priori knowledge that natural millimeter-wave images statistically hold the low-rank property, the proposed approach represents the object images as low-rank matrices and formulates the data acquisition of InSAR in two-dimensional data space directly to undersample visibility function samples. Subsequently, using the undersampled visibility function samples, the optimal solution of the InSAR image reconstruction problem is obtained by simultaneously adopting MC techniques and TV regularization. Experimental results on simulated and real millimeter-wave InSAR image data demonstrate the effectiveness and the significant improvement of the reconstruction performance of the proposed InSAR-TVMC approach over conventional and one-dimensional sparse InSAR image reconstruction approaches.

This paper reports development of a passive millimeter wave imaging system for detection of concealed objects using sparse characterization and reconstruction algorithm based on CS theory. By measuring the radiometric temperatures of different objects, passive millimeter wave imaging can acquire images of objects concealed underneath clothing. In this system a millimeter wave antenna controlled by revolving table receives the passive millimeter signal from target object. As passive millimeter wave images have poor contrast and low signal to noise ratio, traditional segmentation algorithms are unable to detect concealed objects, so this system uses a method which transforms the image signal by a non-adaptive base matrix. The high-dimensional signal will be transformed to a lower dimensional space, and then solve an optimization problem for the highest sparsity of the signal through the gradient projection. We present experimental results of 94GHz passive millimeter wave images.