This work introduces FinGAN, a robust received signal strength (RSS) data generator designed to expand RSS fingerprint datasets. Compared to existing generative adversarial models that either rely on known reference positions (RPs) or depend on predefined priors, FinGAN learns the latent information between RPs and RSS values by maximizing the mutual information between the generated RSS data and the RPs, enabling an end-to-end RSS generation directly from RPs. This allows us to accurately generate RSS data for previously unmeasured RPs. Both quantitative and qualitative evaluations demonstrate that FinGAN produces synthetic RSS data closely aligned with real RSS sample collected from the on-site experiment, preserving localization performance comparable to that achieved with complete real-world datasets. To further validate its generalizability, FinGAN is also trained and evaluated on open-source datasets from three typical office environments,and the results demonstrate consistent performance across different scenarios.

This letter presents a feature-guided adversarial framework, namely ComGAN, which is designed to reconstruct an incomplete fingerprint database by inferring missing received signal strength (RSS) values at unmeasured reference points (RPs). An auxiliary subnetwork is integrated into a conditional generative adversarial network (cGAN) to enable spatial feature learning. An optimization method is then developed to refine the RSS predictions by aggregating multiple prediction sets, achieving an improved localization performance. Experimental results demonstrate that the proposed scheme achieves a root mean squared error (RMSE) comparable to the ground-truth measurements while outperforming state-of-the-art reconstruction methods. When the reconstructed fingerprint is combined with measured data for training, the fingerprinting localization achieves accuracy comparable to models trained on fully measured datasets.

In this work, an end-to-end generative adversarial framework for computational microwave imaging (CMI) is proposed to reconstruct the targets of interest directly from the measurements of targets obstructed by undesired objects. It integrates a conditional generative adversarial network (cGAN) with a learnable soft-threshold module (STM) to adaptively suppress non-target related information. The proposed framework is evaluated on a diverse dataset comprising MNIST digits obstructed by E-MNIST letters for training and testing, as well as on measurements acquired with an experimental CMI system. In addition, further studies involving objects with different geometries are conducted, demonstrating that the proposed approach can be adapted to other types of objects. Numerical experiments show that the proposed cGAN-STM achieves a normalized mean square error (NMSE) of 0.066 and a structural similarity index (SSIM) of 0.876 under ideal conditions. Comprehensive analyses, including benchmarking, analysis of the STM mechanism, and evaluation under different obstruction sizes, are also conducted. The performance of the model under different signal-to-noise ratio (SNR) scenarios is also evaluated, achieving reasonable reconstruction quality at 15 dB SNR with an NMSE of 0.168 and an SSIM of 0.700. Even at low SNR levels, recognizable target outlines are preserved. These results highlight the effectiveness and adaptability of the proposed method.

Through-wall imaging is crucial for applications where identifying and localizing concealed objects behind walls is essential. However, in such scenario, interpreting radar data remains challenging due to complex wall interactions, signal attenuation, and variations in target geometries. In this work, a deep learning-based framework for target classification and localization using B-scan radar images is proposed, where targets are positioned behind walls. The B-scan data, represented as two-dimensional time-trace targets, are used to train neural networks for classifying the presence of targets, while a regressionbased neural network is introduced for estimating the spatial coordinates of targets within the radargrams, enabling effective localization. The proposed models are trained on a dataset comprising simulated B-scan images with varying wall and target configurations. This approach demonstrates robust performance on both the tasks, achieving high accuracy in classification and low localization error. These results highlight the potential of deep learning to automate and enhance the interpretability of B-scan data in non-invasive through-wall radar imaging scenarios.

Although conventional microwave imaging techniques offer high-fidelity image reconstructions, they suffer from significant drawbacks, such as intensive hardware complexity. Computational microwave imaging (CMI)-based systems are often considered alternatives to address these challenges, but they are usually limited by the complexity of the computation layer due to the intensive computational demands for the reconstruction step. A variety of deep learning techniques have been proposed to enhance the computation efficiency, but existing studies primarily focus on scenarios where the imaged targets are isolated. To address this challenge, this work explores computational imaging scenarios involving targets of interest overlapping with secondary objects. This article designs a novel generative model to reconstruct the images of the objects in the foreground of overlapping imaging targets. Performance is evaluated using average normalized mean squared error (NMSE) and structural similarity index (SSIM). While conventional methods require manual removal of background objects for accurate foreground reconstructions in overlapped imaging, the proposed method achieves this by learning features directly from the measured signals, significantly improving the CMI computational efficiency.

While computational microwave imaging (CMI) addresses several limitations of conventional microwave imaging techniques, such as hardware complexity, it is still constrained by substantial computational resources required for image re-construction. This paper presents a convolutional neural network (CNN)-based approach to enhance the computational efficiency of CMI. The proposed network directly computes the transfer function, or sensing matrix, from the aperture fields of antennas within a CMI system. To improve information extraction, convolutional block attention modules (CBAMs) are integrated into the architecture. Numerical results on a testing dataset demonstrate an average normalized mean squared error (NMSE) of 0.054. Compared to conventional methods, the proposed network reduces computation time by 69% in generating the sensing matrix. Overall, the network generates the sensing matrix from two different sets of aperture field distributions with high precision, achieving considerable computational savings for CMI applications.

Computational imaging (CI)-based systems have emerged as a viable alternative to address the challenges of high hardware complexity and slow data acquisition speed associated with conventional microwave imaging. However, CI-based systems are limited by a substantial computational burden during the scene reconstruction process. In particular, image reconstruction and target classification problems for CI systems are computationally complex tasks. To tackle this challenge, a generative deep learning model named ClassiGAN is proposed to jointly solve the image reconstruction and target classification tasks by only using the backscattered measured signals as input. In particular, an adaptive loss function is employed to effectively integrate the respective loss functions for the two tasks, thereby enhancing training efficiency. This adaptive loss function dynamically adjusts the weights of the losses associated with each task, facilitating a more effective integration of the differing loss functions. Notably, ClassiGAN significantly reduces the run time for image reconstruction tasks compared to conventional CI methods. Compared to other state-of-the-art methods, ClassiGAN not only achieves lower average normalized mean squared error (NMSE) and higher structural similarity (SSIM) but also provides a higher accuracy in recognizing imaging targets. Extensive experimental tests further validate ClassiGAN's capability to simultaneously reconstruct and recognize the imaging target within practical settings. Hence, this shows that ClassiGAN can enhance the overall efficiency of CI-based systems at microwave frequencies by addressing challenges related to computational load during run time.

Computational microwave imaging (CMI) has gained attention as an alternative technique for conventional microwave imaging techniques, addressing their limitations such as hardware-intensive physical layer and slow data collection acquisition speed to name a few. Despite these advantages, CMI still encounters notable computational bottlenecks, especially during the image reconstruction stage. In this setting, both image recovery and object classification present significant processing demands. To address these challenges, our previous work introduced ClassiGAN, which is a generative deep learning model designed to simultaneously reconstruct images and classify targets using only back-scattered signals. In this study, we build upon that framework by incorporating attention gate modules into ClassiGAN. These modules are intended to refine feature extraction and improve the identification of relevant information. By dynamically focusing on important features and suppressing irrelevant ones, the attention mechanism enhances the overall model performance. The proposed architecture, named Att-ClassiGAN, significantly reduces the reconstruction time compared to traditional CMI approaches. Furthermore, it outperforms current advanced methods, delivering improved Normalized Mean Squared Error (NMSE), higher Structural Similarity Index (SSIM), and better classification outcomes for the reconstructed targets.

A near-field (NF) measurement method stemming from the microwave computational imaging (MCI) concept is proposed in this letter. In an MCI system, the sensing matrix comprises the radiated fields of both the transmitter and receiver. We demonstrate that, with apriori knowledge of an antenna radiating quasi-random patterns tailored for MCI, the sensing matrix allows for the calculation of the NF distribution of an antenna-under-test (AUT). The proposed technique follows a two-step process. First, the sensing matrix of an MCI system with fixed calibration targets is directly predicted from the back-scattered data by using a trained Pix2pix conditional generative adversarial network (Pix2pix cGAN). Then, with the assistance of the Pix2pix cGAN and MCI-oriented antenna's NF distribution, the NF distribution of the AUT can be retrieved, avoiding time-consuming probe-based NF measurements. The proposed method is validated using various types of antennas exhibiting different polarization states as the AUTs through full-wave simulations.

Conventional microwave imaging can provide high-quality reconstructed images, but is also limited by the increased hardware complexity and a slow data acquisition speed. Although computational imaging (CI)-based systems are developed to be alternatives, they may require substantial computational power and time. To reduce the hardware complexity and computational burden associated with scene reconstructions of CI applications, in this paper, a conditional generative adversarial network (cGAN) is presented to achieve image reconstruction, where the back-scattered measurement is regarded as both the condition and the input of the proposed network. With testing dataset, the average values of the normalized mean squared error (NMSE) and the normalized mean absolute error (NMAE) are 0.0474 and 0.2267, respectively. In addition, a noise analysis is conducted, showing the reliability of the proposed network in noisy settings.

Conventional microwave imaging-based approaches can produce high quality image reconstructions. At the same time, these techniques typically suffer from increased hardware complexity, cost and slow data acquisition speeds. Although computational imaging (CI)-based systems have been developed as an alternative, they may demand significant computational power and time, especially in the calculation and the storage of the transfer function (or the sensing matrix) of the CI system. However, the previous method considers the scenario where the transmitter and receiver share the same set of aperture distribution fields. To address this challenge, this paper presents a new technique, where the sensing matrix is calculated directly from the aperture fields of the antennas in a CI system. Here, the transmitter and the receiver apertures can be different and they do not necessarily need to have the same field distributions. With the testing dataset, the average value of the normalized mean squared error (NMSE) is 0.0243. In addition, compared to the traditional method, the proposed network reduces the computation time for the sensing matrix by approximately 67%. The proposed network can predict the sensing matrix from two different sets of aperture distribution fields with high accuracy while significantly saving the computation time.

This work aims to simplify the characterization process of coded-apertures for computational imaging (CI) at microwave frequencies. A major benefit of the presented technique is the minimization of the processing time needed to calculate the system sensing matrix for microwave CI-based compressive sensing applications. To achieve this, a deep learning-based approach which is capable of generating the sensing matrix using features learned directly from the coded-aperture distribution is proposed. To avoid the vanishing gradient problem, the proposed deep learning network contains skip connections. Using a dataset of 1,000 testing samples, the average normalized mean-squared-error (NMSE) calculated between the sensing matrix generated by the conventional method and that predicted by the proposed network is 0.0036. Moreover, the average mean-squared-error (MSE) calculated between the images reconstructed using the conventional and the predicted sensing matrix is 0.00297. In addition to providing high-fidelity estimations with minimized error, we demonstrate that using the trained network, the prediction of the sensing matrix can be achieved in 0.212 s, corresponding to a 65% reduction in the computation time needed to calculate the sensing matrix. This has significant outcomes in achieving real-time operation of CI-based microwave imaging systems.

In this letter, the challenge of enhancing the efficiency of computational imaging (CI) at microwave frequencies is addressed. While CI simplifies the hardware complexity of conventional microwave imaging techniques, it requires the knowledge of a sensing matrix that is governed by the aperture radiated fields. This can be a computationally expensive process. As a drastic alternative to this conventional approach, a Pix2pix conditional generative adversarial network is introduced to learn the intricate relationship between the back-scattered measurements from the imaging scene and the sensing matrix of the imaging system. The proposed network yields high-fidelity estimations with minimized error and achieves a remarkable reduction in the time required to compute the sensing matrix. This advancement holds significant potential for improving the overall efficiency of microwave CI techniques, addressing both hardware complexity and computational burdens.

Electromagnetic (EM) waves at millimeter-wave (mmW) frequencies have found applications in a variety of imaging systems, from security screening to defense and automotive radars, with the research and development of mmW imaging systems gaining interest in recent years. Despite their significant advantages, mmW imaging systems suffer from poor resolution compared to higher frequency reconstructions, such as optical images. To improve the resolution of mmW images, various super-resolution (SR) techniques have been introduced. One such technique is the use of machine learning algorithms in the signal processing layer of the imaging system without altering any of the system’s parameters. This article focuses on the use of a convolutional neural network (CNN) architecture to achieve SR when applied to 3-D mmW input images. To exploit the phase information content of the input images along with the magnitude, a complex-valued CNN is designed, which can accommodate complex-valued data. To simplify the learning process, the resolution difference between the input and output images is divided into smaller parts by using subnetworks in the CNN architecture. The trained model is tested on simulated and experimental targets. The average mean square error score and the structural similarity index obtained on a test dataset of 460 samples are 0.0127 and 0.9225, respectively. It can be inferred that the model has the capability to improve the resolution of input mmW images to a high degree of fidelity, hence paving the way for an end-to-end SR imaging system.