Compressive Imaging Research Articles

Nondestructive testing of GFRP composites with compressive sensing-based terahertz single-pixel imaging

ABSTRACT Terahertz (THz) imaging technique has a wide range of applications, from industrial non-destructive testing (NDT) to various biomedical applications. Although the capabilities of terahertz radiation in defect detection are well-proven, the practical limitations have been associated with the long image acquisition time. Hence, to improve the image acquisition speed, the compressive sensing technique has been employed. However, the challenge is identifying the optimal modulation mask for compressive sensing as the image reconstruction metrics hugely depend on the mask. Hence, a systematic investigation of the different modulation masks has been done to reconstruct THz images of glass fibre-reinforced polymer (GFRP) composites using TVAL3 (Total Variation minimization by Augmented Lagrangian and ALternating direction Algorithm). The image reconstruction quality has been studied using mean square error, peak signal-to-noise ratio, and structural similarity index. From the results, it can be noticed with a 0.3 sampling ratio, reliable reconstruction of the THz image is possible, saving 70% of the image acquisition time. Further, the discrete cosine transform (DCT) mask is ideal for high frame rate NDT in low and medium noise scenarios. However, the Bernoulli basis offers high resistance to noise by resulting in the best image reconstruction results at high noise cases, outperforming the DCT.

High-resolution compressive 3D imaging through phase ranging with a single-pixel photodetector.

Single-pixel three-dimensional (3D) imaging is of great potential to the light detection and ranging (Lidar) system. Combining high-precision ranging with compressed sensing (CS) imaging enables the acquisition of high-resolution and accurate depth maps under non-scanning conditions. In this study, we present that a phase ranging system can be combined perfectly with the CS algorithm perfectly for real 3D imaging through only two reconstructions, along with high accuracy and fast depth measurement. A structured detection architecture based on amplitude-modulated continuous-wave laser illumination and single-pixel detector with high-precision 3D image reconstruction at 128 × 128 resolution using only a 5% sampling rate were used. The minimum identifiable target width of this system is 0.2 cm, corresponding to a resolving angle of 0.33 mrad. This system paves the reality of high-precision and fast 3D imaging for continuous-wave Lidar detections.

Glow Discharge Optical Emission Coded Aperture Spectroscopy.

Glow discharge optical emission spectrometry (GDOES) allows fast and simultaneous multielemental analysis directly from solids and depth profiling down to the nanometer scale, which is critical for thin-film (TF) characterization. Nevertheless, operating conditions for the best limits of detection (LODs) are compromised in lieu of the best sputtering crater shapes for depth resolution. In addition, the fast transient signals from ultra-TFs do not permit the optimal sampling statistics of bulk analysis such that LODs are further compromised. Furthermore, commercial GDOES instruments rely on a slit-based light dispersion that favors high spectral resolution at the expense of light throughput. Here, a new technique called glow discharge optical emission coded aperture spectrometry (GOCAS) is shown to allow both a higher spectral resolution and higher light throughput by using a coded aperture (CA) with multiple thin slits at the spectrograph's entrance to measure the convoluted spectra and compressed sensing (CS) algorithms to recover the deconvoluted spectra from the full field of view. The effects of CA characteristics on spectral reconstruction fidelity were studied and showed the best fidelity for smaller slits, 50% transmittance, and wider CA with a higher number of slits. In addition, Shearlet enhanced snapshot compressive imaging (SeSCI)GPU showed the best performance of the CS algorithms studied, including SeSCICPU, two-step iterative shrinkage/thresholding (TwIST), and alternating direction method of multipliers total variation minimization (ADMM-TV). Moreover, GOCAS is shown to be very robust against increasing detector Gaussian noise. Finally, standard reference materials are used to show up to ∼30× improved S/N and an order-of-magnitude improved LODs, at the fastest acquisition times (fraction of a ms), which has the potential to be transformative for depth profiling of nanostructured materials.

An Encoded Photodetector Array for High-Frame-Rate Compressed Sensing and Computational Imaging

An Encoded Photodetector Array for High-Frame-Rate Compressed Sensing and Computational Imaging

Single-shot Compressive Hyperspectral Imaging Utilizing Both Positive and Negative Diffracted Waves

Single-shot Compressive Hyperspectral Imaging Utilizing Both Positive and Negative Diffracted Waves

Compressive electron backscatter diffraction imaging.

Electron backscatter diffraction (EBSD) has developed over the last few decades into a valuable crystallographic characterisation method for a wide range of sample types. Despite these advances, issues such as the complexity of sample preparation, relatively slow acquisition, and damage in beam-sensitive samples, still limit the quantity and quality of interpretable data that can be obtained. To mitigate these issues, here we propose a method based on the subsampling of probe positions and subsequent reconstruction of an incomplete data set. The missing probe locations (or pixels in the image) are recovered via an inpainting process using a dictionary-learning based method called beta-process factor analysis (BPFA). To investigate the robustness of both our inpainting method and Hough-based indexing, we simulate subsampled and noisy EBSD data sets from a real fully sampled Ni-superalloy data set for different subsampling ratios of probe positions using both Gaussian and Poisson noise models. We find that zero solution pixel detection (inpainting un-indexed pixels) enables higher-quality reconstructions to be obtained. Numerical tests confirm high-quality reconstruction of band contrast and inverse pole figuremaps from only 10% of the probe positions, with the potential to reduce this to 5% if only inverse pole figuremaps are needed. These results show the potential application of this method in EBSD, allowing for faster analysis and extending the use of this technique to beam sensitive materials.

Block-based compressive imaging with swin transformer

Block-based compressive imaging with swin transformer

A Compressive Super-resolution Imaging Algorithm for Multi-frequency Near-field Millimeter-wave

A Compressive Super-resolution Imaging Algorithm for Multi-frequency Near-field Millimeter-wave

USB-Net: Unfolding Split Bregman Method With Multi-Phase Feature Integration for Compressive Imaging

USB-Net: Unfolding Split Bregman Method With Multi-Phase Feature Integration for Compressive Imaging

Anti-perturbation high-resolution compressive imaging through flexible fibers via random-phased fiber plates

Anti-perturbation high-resolution compressive imaging through flexible fibers via random-phased fiber plates

Dynamic light field reconstruction via densely connected deep equilibrium model

High-resolution consumer plenoptic cameras usually feature low frame rates, making them not well-suited for capturing high-speed motion scenes. To compensate for this limitation, we extend the original snapshot compressive imaging system to plenoptic cameras and propose a densely connected deep equilibrium (DEQ) model for high-quality dynamic light field (LF) reconstruction, abbreviated as DLFDEQ. Specifically, we perform temporal compression encoding on a dynamic LF and model the reconstruction process as an inverse problem with an implicit regularization term. To solve this inverse problem, we present a densely connected DEQ model based on gradient descent. Our approach demonstrates stronger robustness and better detail retention than existing methods. We can practically quadruple the original camera’s frame rate by continually capturing and retrieving these measurement frames with high reconstruction accuracy.

Motion-Aware Dynamic Graph Neural Network for Video Compressive Sensing.

Video snapshot compressive imaging (SCI) utilizes a 2D detector to capture sequential video frames and compress them into a single measurement. Various reconstruction methods have been developed to recover the high-speed video frames from the snapshot measurement. However, most existing reconstruction methods are incapable of efficiently capturing long-range spatial and temporal dependencies, which are critical for video processing. In this paper, we propose a flexible and robust approach based on the graph neural network (GNN) to efficiently model non-local interactions between pixels in space and time regardless of the distance. Specifically, we develop a motion-aware dynamic GNN for better video representation, i.e., represent each node as the aggregation of relative neighbors under the guidance of frame-by-frame motions, which consists of motion-aware dynamic sampling, cross-scale node sampling, global knowledge integration, and graph aggregation. Extensive results on both simulation and real data demonstrate both the effectiveness and efficiency of the proposed approach, and the visualization illustrates the intrinsic dynamic sampling operations of our proposed model for boosting the video SCI reconstruction results. The code and model will be released.

Improved fast deconvolution algorithms based on functional beamforming for gas leakage sound source imaging

Fast deconvolution algorithms have high computational efficiency, but they often overlook spatial variations in the point spread function (PSF) and wrap-around errors, resulting in the degradation of sound source imaging performance. Accordingly, this paper proposes two improved fast deconvolution algorithms based on functional beamforming (FB). The proposed algorithms establish a linear equation involving the FB output, sound source distribution, and raised-power spatial shift-invariant PSF, specially with irregular focus grid point. Furthermore, iterative solutions are obtained based on fast Fourier transform, enhancing imaging performance while maintaining computational efficiency. Simulations and experimental results demonstrate that compared to the fast deconvolution algorithms, the proposed algorithms expand the effective imaging area to 36.3°, and the Rényi entropy is reduced by approximately 45 %, which has higher imaging spatial resolution and dynamic range. Finally, the compressed gas leakage source imaging results validate the effectiveness of the proposed algorithms, indicating that they have promising engineering application prospects.

Mask-Guided Spatial-Spectral MLP Network for High-Resolution Hyperspectral Image Reconstruction.

Hyperspectral image (HSI) reconstruction is a critical and indispensable step in spectral compressive imaging (CASSI) systems and directly affects our ability to capture high-quality images in dynamic environments. Recent research has increasingly focused on deep unfolding frameworks for HSI reconstruction, showing notable progress. However, these approaches have to break the optimization task into two sub-problems, solving them iteratively over multiple stages, which leads to large models and high computational overheads. This study presents a simple yet effective method that passes the degradation information (sensing mask) through a deep learning network to disentangle the degradation and the latent target's representations. Specifically, we design a lightweight MLP block to capture non-local similarities and long-range dependencies across both spatial and spectral domains, and investigate an attention-based mask modelling module to achieve the spatial-spectral-adaptive degradation representationthat is fed to the MLP-based network. To enhance the information flow between MLP blocks, we introduce a multi-level fusion module and apply reconstruction heads to different MLP features for deeper supervision. Additionally, we combine the projection loss from compressive measurements with reconstruction loss to create a dual-domain loss, ensuring consistent optical detection during HS reconstruction. Experiments on benchmark HS datasets show that our method outperforms state-of-the-art approaches in terms of both reconstruction accuracy and efficiency, reducing computational and memory costs.

Toward video-rate compressive spontaneous Raman imaging via single-photon avalanche diode arrays.

Spontaneous Raman microscopy is well-known for its remarkable chemical contrast yet suffers from slow acquisition speeds. Recently, the compressive Raman microspectroscopy framework has shown that a significant speed advantage is brought by leveraging shot-noise-limited detection using a single-photon avalanche diode (SPAD). However, current imaging speeds of compressive Raman architectures are fundamentally limited by SPAD sensitivity and dead time. Here, we demonstrate an efficient and scalable compressive Raman parallelization scheme based on SPAD arrays. We show that parallelization using line excitation, instead of spatial multiplexing, allows to reach effective pixel dwell times (τ pdt ) of 0.8 µs. Such fast speed represents over one order-of-magnitude speed-up over previous demonstrations. This effective parallelization not only allows for demonstrating unprecedented chemical imaging speeds using the otherwise weak spontaneous Raman effect but also paves the way for true video-rate inexpensive molecular microspectroscopy.

Combining Low-Rank and Deep Plug-and-Play Priors for Snapshot Compressive Imaging.

Snapshot compressive imaging (SCI) is a promising technique that captures a 3-D hyperspectral image (HSI) by a 2-D detector in a compressed manner. The ill-posed inverse process of reconstructing the HSI from their corresponding 2-D measurements is challenging. However, current approaches either neglect the underlying characteristics, such as high spectral correlation, or demand abundant training datasets, resulting in an inadequate balance among performance, generalizability, and interpretability. To address these challenges, in this article, we propose a novel approach called LR2DP that integrates the model-driven low-rank prior and data-driven deep priors for SCI reconstruction. This approach not only captures the spectral correlation and deep spatial features of HSI but also takes advantage of both model-based and learning-based methods without requiring any extra training datasets. Specifically, to preserve the strong spectral correlation of the HSI effectively, we propose that the HSI lies in a low-rank subspace, thereby transforming the problem of reconstructing the HSI into estimating the spectral basis and spatial representation coefficient. Inspired by the mutual promotion of unsupervised deep image prior (DIP) and trained deep denoising prior (DDP), we integrate the unsupervised network and pre-trained deep denoiser into the plug-and-play (PnP) regime to estimate the representation coefficient together, aiming to explore the internal target image prior (learned by DIP) and the external training image prior (depicted by pre-trained DDP) of the HSI. An effective half-quadratic splitting (HQS) technique is employed to optimize the proposed HSI reconstruction model. Extensive experiments on both simulated and real datasets demonstrate the superiority of the proposed method over the state-of-the-art approaches.

Noise-robust and data-efficient compressed ghost imaging via the preconditioned S-matrix method.

The design of the illumination pattern is crucial for improving imaging quality of ghost imaging (GI). The S-matrix is an ideal binary matrix for use in GI with non-visible light and other particles since there are no uniformly configurable beam-shaping modulators in these GI regimes. However, unlike widely researched GI with visible light, there is relatively little research on the sampling rate and noise resistance of compressed GI based on the S-matrix. In this paper, we investigate the performance of compressed GI using the S-matrix as the illumination pattern (SCSGI) and propose a post-processing method called preconditioned S-matrix compressed GI (PSCSGI) to improve the imaging quality and data efficiency of SCSGI. Simulation and experimental results demonstrate that compared with SCSGI, PSCSGI can improve imaging quality in noisy conditions while utilizing only half the amount of data used in SCSGI. Furthermore, better reconstructed results can be obtained even when the sampling rate is as low as 5%. The proposed PSCSGI method is expected to advance the application of binary masks based on the S-matrix in GI.

Real-Time Decoding of Snapshot Compressive Imaging Using Tensor FISTA-Net.

Snapshot compressive imaging (SCI) cameras compress high-speed videos or hyperspectral images into measurement frames. However, decoding the data frames from measurement frames is compute-intensive. Existing state-of-the-art decoding algorithms suffer from low decoding quality or heavy running time or both, which are not practical for real-time applications. In this article, we exploit the powerful learning ability of deep neural networks (DNN) and propose a novel tensor fast iterative shrinkage-thresholding algorithm net (Tensor FISTA-Net) as a real-time decoder for SCI cameras. Since SCI cameras have an accurate physical model, we can trade training time for the decoding time by generating abundant synthetic data and training a decoder on the cloud. Tensor FISTA-Net not only learns a sparse representation of the frames through convolution layers but also reduces the decoding time and memory consumption significantly through tensor operations, which makes Tensor FISTA-Net an appropriate approach for a real-time decoder. Our proposed Tensor FISTA-Net obtains an average PSNR improvement of 0.79-2.84 dB (video images) and 2.61-4.43 dB (hyperspectral images) over the state-of-the-art algorithms, along with more clear and detailed visual results on real SCI datasets, Hammer and Wheel, respectively. Our Tensor FISTA-Net reaches 45 frames per second in video datasets and 70 frames per second in hyperspectral datasets, meeting the real-time requirement. Besides, the trained model occupies only a 12 -MB memory footprint, making it applicable to real-time Internet of Things (IoT) applications.

Dual-Domain Feature Fusion and Multi-Level Memory-Enhanced Network for Spectral Compressive Imaging

Dual-Domain Feature Fusion and Multi-Level Memory-Enhanced Network for Spectral Compressive Imaging

Wavefront shaping and imaging through a multimode hollow-core fiber.

Multimode fibers recently emerged as compact minimally-invasive probes for high-resolution deep-tissue imaging. However, the commonly used silica fibers have a relatively low numerical aperture (NA) limiting the spatial resolution of a probe. On top of that, light propagation within the solid core generates auto-fluorescence and Raman background, which interferes with imaging. Here we propose to use a hollow-core fiber to solve these problems. We experimentally demonstrate spatial wavefront shaping at the multimode hollow-core fiber output with tunable high-NA. We demonstrate raster-scan and speckle-based compressive imaging through a multimode hollow-core fiber.