Resolution Enhancement for Millimeter-Wave Radar ROI Image with Bayesian Compressive Sensing

Pengfei Xie, Jianxin Wu, Lei Zhang, Guanyong Wang, Jin XueSensors 2022 · DOI

TL;DR

This paper introduces a novel method for enhancing the resolution of millimeter-wave (MMW) radar images, specifically focusing on regions of interest (ROIs) containing suspicious targets. By leveraging the spatial sparsity inherent in MMW images and employing Bayesian Compressive Sensing (BCS) with Fast Fourier Transforms (FFT), the authors achieve significant resolution improvements without requiring hardware modifications, making it a practical solution for security applications.

What Problem Does This Paper Address?

Millimeter-wave (MMW) radar imaging is crucial for security applications, such as concealed weapon detection, due to its ability to penetrate materials and provide high-resolution images. However, conventional MMW imaging techniques, particularly synthetic aperture radar (SAR), often suffer from limited resolution dictated by system parameters. This limitation hinders the accurate detection and recognition of small or subtle suspicious targets. While existing super-resolution (SR) methods exist, they often necessitate costly hardware upgrades or complex system modifications. This paper aims to address the challenge of improving MMW image resolution, especially for specific regions of interest (ROIs) where threats are likely to be found, without altering the imaging hardware.

Methods

The core of this paper's contribution lies in its application of Bayesian Compressive Sensing (BCS) for super-resolution (SR) of MMW images. The authors exploit the inherent spatial sparsity of MMW images, particularly in regions of interest (ROIs), which means that only a few pixels contain significant information. BCS is chosen because it can effectively utilize this prior knowledge of sparsity to reconstruct a higher-resolution image from fewer measurements.

The process involves:

  1. Holographic Imaging and Wave Number Spectrum Analysis: The paper first details the active MMW 3D holographic imaging system and analyzes the characteristics of its spatial wave number spectrum. This analysis helps understand the limitations and potential for resolution enhancement.
  2. MMW 3D Image Sparsity and ROI Selection: The authors demonstrate that MMW images, especially 2D projections, exhibit sparsity. They propose focusing on a Region of Interest (ROI) identified in a lower-resolution projection image. This selective processing significantly reduces computational load.
  3. Bayesian Compressive Sensing (BCS) for SR: For the selected ROI, the paper applies BCS. The BCS algorithm is formulated as a sparse reconstruction problem, leveraging a 2D Fourier transform matrix as the compressive sensing matrix. This choice is motivated by the computational efficiency of using FFT for implementation and the ability to analyze the wave number spectrum. The BCS approach uses a probabilistic framework (Bayesian learning) to estimate the high-resolution image by maximizing the posterior probability, effectively incorporating the sparsity prior.
  4. Regional 3D Image SR: To further optimize the process, a regional 3D SR technique is developed. This involves processing only the relevant 2D slices of the 3D MMW image corresponding to the ROI, rather than the entire 3D volume.

Datasets

The paper utilizes both simulated and measured data for evaluation.

  • Simulated Data: 2D lattice point targets were used to simulate MMW imaging scenarios. This allowed for controlled experiments to evaluate the SR algorithm's performance on known structures.
  • Measured Data: Real MMW 3D imaging data of a human body was collected. This dataset included hidden objects (a toy gun and a USB disk) to demonstrate the practical application of the SR method in a security context.

Key Insights and Ideas

  • Sparsity is Key: The paper strongly emphasizes that the spatial sparsity of MMW images is a critical property that can be exploited for effective super-resolution without hardware changes.
  • ROI Focus for Efficiency: By concentrating SR processing on identified Regions of Interest (ROIs), computational burden is significantly reduced, making the method practical for real-time security applications.
  • BCS for Robust Reconstruction: Bayesian Compressive Sensing offers a robust framework for SR by incorporating prior knowledge about the image's sparsity, leading to improved resolution and reduced artifacts.
  • FFT for Computational Speed: The use of FFT as the compressive sensing matrix is a clever choice that leverages existing efficient algorithms for faster computation.
  • Hardware-Agnostic Improvement: The proposed method offers a significant advantage by enhancing image quality without requiring expensive hardware upgrades, making it a cost-effective solution.

Key Results

  • Resolution Enhancement: The proposed BCS-based SR method demonstrably improved the resolution of MMW images. For lattice point targets, the 3 dB resolution in the X dimension improved from 3.9 mm to 1.8 mm, and in the Z dimension from 5.6 mm to 2.7 mm.
  • Side-Lobe Reduction: The SR process significantly reduced side-lobe amplitudes by approximately 40 dB, leading to cleaner images and better target separation.
  • Practical Application: In the human body imaging experiment, the resolution in the X-Z plane increased from approximately 5.5 mm to 4 mm after applying the regional SR method.
  • Improved Image Clarity: Qualitative results showed that the SR images were clearer, with better-defined edges and details compared to images processed with bilinear interpolation.

Limitations and Future Work

  • Computational Cost: While the ROI approach reduces computation, BCS can still be computationally intensive for very large ROIs or real-time processing on resource-constrained devices. The paper mentions that FFT speeds up computation, but further optimization might be needed.
  • Sparsity Assumption: The effectiveness of BCS relies heavily on the assumption of sparsity. If the target or background in the ROI is not sufficiently sparse, the SR performance might degrade.
  • Parameter Tuning: The performance of BCS can be sensitive to parameter tuning (e.g., the sparse coefficient $\mu$).
  • Future Work: The paper suggests that this regional SR technique can be further applied to improve threat detection and recognition accuracy. Future work could explore adaptive ROI selection, more advanced sparsity priors, or real-time implementation on embedded systems.

Summary updated

Abstract

For millimeter-wave (MMW) imaging security systems, the image resolution promisingly determines the performance of suspicious target detection and recognition. Conventional synthetic aperture radar (SAR) imaging algorithms only provide limited resolution in active MMW imaging, which is limited by the system. In terms of enhancing the resolution of a region of interest (ROI) image containing suspicious targets, super-resolution (SR) imaging is adopted via Bayesian compressive sensing (BCS) implemented by fast Fourier transform (FFT). The spatial sparsity of MMW ROI images is well exploited with BCS to achieve resolution enhancement without computational cost. Both simulated and measured experiments confirm that the proposed scheme effectively improves the resolution of ROI images.
BibTeX
@inproceedings{xie2022resolution,
  author = {Pengfei Xie and Jianxin Wu and Lei Zhang and Guanyong Wang and Jin Xue},
  title = {{Resolution Enhancement for Millimeter-Wave Radar ROI Image with Bayesian Compressive Sensing}},
  booktitle = {Sensors},
  year = {2022},
  doi = {10.3390/s22155757},
}
Figure