3-D Super-Resolution of Coded Aperture Millimeter-Wave Images Using Complex-Valued Convolutional Neural Network

Abstract

Electromagnetic waves at millimetre-wave (mmW) frequencies have found applications in a variety of imaging systems, from security screening to defence 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 as 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 paper focuses on the use of a convolutional neural network (CNN) architecture to achieve SR when applied to three-dimensional mmW input images. To exploit the phase information content of the input images along with the magnitude, a complex-valued CNN is designed that can accommodate complexvalued data. To simplify the learning process, the resolution difference between the input and output images is divided into smaller parts by using sub-networks in the CNN architecture. The trained model is tested on simulated as well as experimental targets. The average mean-squared-error score and 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.

Super-resolution Reconstruction and Denoising of 3D Millimetre-wave Images using a Complex-valued Convolutional Neural Network

“…To validate the developed CV-CNN model with experimental reconstruction data, we test the performance of the CV-CNN model with images reconstructed from an experimental system. The measurement of this study is carried out using a SAR imaging set-up in a near-field anechoic chamber [10], [18] and at a frequency band of 18-20 GHz. This system is used to image a target consisting of an array of copper stripes along the horizontal and vertical axes, as shown in Fig.…”

“…The architecture selected for the CV-CNN to solve the SR problem is based on the one developed in [10], as shown in Fig. 1.…”

“…For each noise level, the model is trained for 700 epochs with an initial learning rate of 10 −3 . With every epoch, the learning rate decays by a factor of δ epoch/∆ , where δ is the decay rate which is chosen to be 0.96, epoch is the current epoch and ∆ is the decay step which is chosen to be 1000 [10]. Choosing an optimal decay step is important as allowing the learning rate to decay to a small value can make the training process slow and the network will take a longer time to converge.…”

Single Radar Image Motion Deblurring Through PSF Estimation and Convolution Regularizer

“…Results showcased significant advancements over conventional methodologies by effectively eliminating artifacts present in the input images. Our prior work, [48], involved constructing a complex-valued CNN-based super-resolution model tailored for radar images. This model, leveraging three-dimensional (3D) complexvalued CNN architecture, distinguished itself by employing smaller filter sizes and fewer layers, thereby reducing training parameters and time.…”

Axial super-resolution optical coherence tomography via complex-valued network

“…Furthermore, complex-valued networks are advantageous in easy optimization, fast learning, and noise-robust memory mechanisms (Nitta 2002, Danihelka et al 2016, Wisdom et al 2016. It has been successfully applied to achieve super-resolution for Terahertz imaging (Wang et al 2021), hyperspectral imaging (Aburaed et al 2023), and millimeter-wave imaging (Sharma et al 2022).…”