Fast Reconstruction Speed Research Articles

Deep-Learning-Based Real-Time Passive Non-Line-of-Sight Imaging for Room-Scale Scenes.

Non-line-of-sight imaging is a technique for reconstructing scenes behind obstacles. We report a real-time passive non-line-of-sight (NLOS) imaging method for room-scale hidden scenes, which can be applied to smart home security monitoring sensing systems and indoor fast fuzzy navigation and positioning under the premise of protecting privacy. An unseen scene encoding enhancement network (USEEN) for hidden scene reconstruction is proposed, which is a convolutional neural network designed for NLOS imaging. The network is robust to ambient light interference conditions on diffuse reflective surfaces and maintains a fast reconstruction speed of 12.2 milliseconds per estimation. The consistency of the mean square error (MSE) is verified, and the peak signal-to-noise ratio (PSNR) values of 19.21 dB, 15.86 dB, and 13.62 dB are obtained for the training, validation, and test datasets, respectively. The average values of the structural similarity index (SSIM) are 0.83, 0.68, and 0.59, respectively, and are compared and discussed with the corresponding indicators of the other two models. The sensing system built using this method will show application potential in many fields that require accurate and real-time NLOS imaging, especially smart home security systems in room-scale scenes.

Fast reconstruction of EEG signal compression sensing based on deep learning

When traditional EEG signals are collected based on the Nyquist theorem, long-time recordings of EEG signals will produce a large amount of data. At the same time, limited bandwidth, end-to-end delay, and memory space will bring great pressure on the effective transmission of data. The birth of compressed sensing alleviates this transmission pressure. However, using an iterative compressed sensing reconstruction algorithm for EEG signal reconstruction faces complex calculation problems and slow data processing speed, limiting the application of compressed sensing in EEG signal rapid monitoring systems. As such, this paper presents a non-iterative and fast algorithm for reconstructing EEG signals using compressed sensing and deep learning techniques. This algorithm uses the improved residual network model, extracts the feature information of the EEG signal by one-dimensional dilated convolution, directly learns the nonlinear mapping relationship between the measured value and the original signal, and can quickly and accurately reconstruct the EEG signal. The method proposed in this paper has been verified by simulation on the open BCI contest dataset. Overall, it is proved that the proposed method has higher reconstruction accuracy and faster reconstruction speed than the traditional CS reconstruction algorithm and the existing deep learning reconstruction algorithm. In addition, it can realize the rapid reconstruction of EEG signals.

CT image reconstruction via industrial CT fast scanning

In automated manufacturing and safety inspection, there is a high demand for fast computed tomography (CT) scanning and image reconstruction. Currently, faster scanning can be achieved by reducing the X-ray exposure time within sparse view CT. The faster scanning strategy introduces significant streak artefacts and noise during the sampling process. Consequently, streak artefacts and noise need to be simultaneously suppressed, which is poses a challenge for existing reconstruction methods. This paper presents a fast iterative reconstruction algorithm that can simultaneously suppress both streak artefacts and noise. This method can not only reconstruct high-fidelity images from rapidly acquired projection data, but also has a faster reconstruction speed than the existing iterative reconstruction algorithms. First, we present a high-order multi-directional total variation (HOM-TV) method that specifically focuses on preserving edge details of the image. Then, we present a fast iterative reconstruction model by incorporating HOM-TV and non-local means into the objective function. Finally, the effectiveness of the presented reconstruction model is validated by simulation and real experiments. The faster scanning method can complete the scan in only 5 seconds, and the structural similarity index (SSIM) of the CT image reconstructed by our method is 0.9755, which is higher than 0.0175 of the Fast Null Space Reconstruction (FNSR) algorithm. The peak signal-to-noise ratio (PSNR) index is 1.656, which is higher than that of the contrast algorithm. In terms of reconstruction time, our algorithm can achieve reconstruction in as little as 36 seconds, outperforming the baseline algorithms.

A general parametric finite element modeling and stress response analysis methods for pipeline system using the nonconforming solid element

In the dynamic topology optimization design of pipeline system, parametric finite element modeling (PFEM) is needed to improve the model reconstruction efficiency. Taking a typical spatial single pipeline as the research object, this paper innovatively presents a PFEM method of pipeline system based on the newly constructed nonconforming solid (Solid-NC) element. The stiffness and mass matrices of the spatial 8-node Solid-NC element with second-order boundary accuracy are obtained by introducing the node-free displacement items and statics condensation method. The PFEM method of the pipeline body is proposed by selecting the straight-line segment lengths as the pipeline shape control parameters. The PFEM method of the pipeline is described in detail, including the node coordinate solutions of straight-line and curved arc segments based on the direction vector method and vector decomposition method, respectively, and the simulation of the clamp mechanical properties considering the actual pre-tightening state of the clamp. Furthermore, the modal analysis is carried out, the node stress response solution is solved based on the stress smoothing method, and the effectiveness of the PFEM method is verified through experiment. Finally, the mesh quality and model reconstruction efficiency of the PFEM method is analyzed, the results show that the developed PFEM method has higher mesh quality and fast model reconstruction speed, and the model reconstruction time is shortened by 544.5 times at most compared with ANSYS software.

A Method of Printmaking Image Generation Based on Convolutional Neural Network

In this paper, a novel image reconstruction method for printmaking is proposed via the combination of nonlinear diffusion filtering and convolution neural network. Nonlinear diffusion filtering based on a partial differential equation abstracts the input image to extract color features and texture features of images by nonlinear structured tensors. The image color features and texture features are combined to obtain a feature vector, and the feature vector is fed into the model as the input data. Then, the pre-trained deep learning model VGG-19 is utilized as the backbone network for further feature representation, and the extracted feature maps of each layer of the VGG19 model can be visualized for image reconstruction. The quality of the reconstructed images of prints is improved with the powerful feature extraction and combination ability of neural networks, which has the advantages of fast reconstruction speed and high quality of reconstructed images. Finally, the simulation experiments can demonstrate that the proposed reconstruction approach can achieve relatively ideal reconstruction performance in scenes of printmaking.

SALSA-Net: Explainable Deep Unrolling Networks for Compressed Sensing.

Deep unrolling networks (DUNs) have emerged as a promising approach for solving compressed sensing (CS) problems due to their superior explainability, speed, and performance compared to classical deep network models. However, the CS performance in terms of efficiency and accuracy remains a principal challenge for approaching further improvements. In this paper, we propose a novel deep unrolling model, SALSA-Net, to solve the image CS problem. The network architecture of SALSA-Net is inspired by unrolling and truncating the split augmented Lagrangian shrinkage algorithm (SALSA) which is used to solve sparsity-induced CS reconstruction problems. SALSA-Net inherits the interpretability of the SALSA algorithm while incorporating the learning ability and fast reconstruction speed of deep neural networks. By converting the SALSA algorithm into a deep network structure, SALSA-Net consists of a gradient update module, a threshold denoising module, and an auxiliary update module. All parameters, including the shrinkage thresholds and gradient steps, are optimized through end-to-end learning and are subject to forward constraints to ensure faster convergence. Furthermore, we introduce learned sampling to replace traditional sampling methods so that the sampling matrix can better preserve the feature information of the original signal and improve sampling efficiency. Experimental results demonstrate that SALSA-Net achieves significant reconstruction performance compared to state-of-the-art methods while inheriting the advantages of explainable recovery and high speed from the DUNs paradigm.

Improved block sparse Bayesian learning based DOA estimation for massive MIMO systems

Direction-of-arrival (DOA) estimation is crucial for suppressing inter-cell interference in massive multiple-input multiple-output (MIMO) systems. However, conventional algorithms are unsuitable for low signal-to-noise ratio (SNR) and a small number of snapshots. Consequently, in an attempt to achieve fast reconstruction speed while ensuring high accuracy, this study proposes an improved block sparse Bayesian learning-based DOA estimation algorithm for two-dimension massive MIMO systems. Using the sparse characteristics and inherent structure of the signal, the DOA estimation problem is transformed to a joint task of sparse signal reconstruction and parameter optimization. Subsequently, the estimation accuracy is improved through complete excavation and learning of the signal temporal structure. In addition, to avoid the direct solution of the posterior probability distribution, reduce the number of iterations and increase the reconstruction speed, variational inference is used to approximate the Bayesian model and solve the joint problem through parameter learning. Moreover, the auxiliary angle is introduced to avoid extra angle matching, which further reduces the complexity. Simulations are performed to verify the proposed method and the results prove its superiority when applied to massive MIMO systems compared with some existing DOA estimation algorithms.

Joint Deep Learning and Information Propagation for Fast 3D City Modeling

In the field of geoinformation science, multiview, image-based 3D city modeling has developed rapidly, and image depth estimation is an important step in it. To address the problems of the poor adaptability of training models of existing neural network methods and the long reconstruction time of traditional geometric methods, we propose a general depth estimation method for fast 3D city modeling that combines prior knowledge and information propagation. First, the original image is downsampled and input into the neural network to predict the initial depth value. Then, depth plane fitting and joint optimization are combined with the superpixel information and the superpixel optimized depth value is upsampled to the original resolution. Finally, the depth information propagation is checked pixel-by-pixel to obtain the final depth estimate. Experiments were conducted using multiple image datasets taken from actual indoor and outdoor scenes. Our method was compared and analyzed with a variety of existing widely used methods. The experimental results show that our method maintains high reconstruction accuracy and a fast reconstruction speed, and it achieves better performance. This paper offers a framework to integrate neural networks and traditional geometric methods, which provide a new approach for obtaining geographic information and fast 3D city modeling.

Artificial Intelligence-Based Image Reconstruction for Computed Tomography: A Survey

Computed tomography has made significant advances since its introduction in the early 1970s, where researchers have mainly focused on the quality of image reconstruction in the early stage. However, radiation exposure poses a health risk, prompting the demand of the lowest possible dose when carrying out CT examinations. To acquire high-quality reconstruction images with low dose radiation, CT reconstruction techniques have evolved from conventional reconstruction such as analytical and iterative reconstruction, to reconstruction methods based on artificial intelligence (AI). All these efforts are devoted to constructing high-quality images using only low doses with fast reconstruction speed. In particular, conventional reconstruction methods usually optimize one aspect, while AI-based reconstruction has finally managed to attain all goals in one shot. However, there are limitations such as the requirements on large datasets, unstable performance, and weak generalizability in AI-based reconstruction methods. This work presents the review and discussion on the classification, the commercial use, the advantages, and the limitations of AI-based image reconstruction methods in CT.

A camera array based reconstruction method for limited observation windows projection in three-dimensional flame chemiluminescence tomography

Three-dimensional(3D) flame chemiluminescence tomography (FCT) is realized in multi-direction and multi-angle using traditional iterative algorithms. However, 3D-FCT is often challenged by insufficient sampling data because of the limited observation windows provided in practical for flame measurement. To obtain flame multiple projections in limited observation windows, we develop a camera array arrangement in FCT. In addition, a residual network with a new loss function combing physical model of flame is proposed to achieve higher reconstruction accuracy, faster reconstruction speed and efficient image feature in residual network for 3D FCT measurement. Furthermore, the determination of the weight coefficient in the loss function is performed by numerical simulation experiments. The flame reconstruction results show that the proposed residual network method including the new loss function has more reliable structural similarity and noise immunity compared with the ART algorithm and the CNN algorithm. This work provides a faster and more accurate method for combustion diagnosis under limited observation windows with insufficient projections.

Reconstruction of tomographic gamma scanning transmission image from sparse projections based on convolutional neural networks

Tomographic Gamma Scanning (TGS) is an important Non-Destructive Assay (NDA) method for radioactive waste, and transmission imaging is the basis of the TGS. Traditional algorithms reconstruct the transmission image require the provision of the same amount of projection data as the pixel value of the transmission image, which requires a long time to scan and severely limits the industrial application of TGS. Sparse angle scanning is an optimal approach to improve the efficiency of the TGS system, but the amount of data generated by sparse angle scanning is difficult to support traditional algorithms, and blurs and artifacts will appear in the reconstructed image. In this work, Algebraic Reconstruction Technique (ART) combined with the Residual Network(ResNet) to reconstruct TGS transmission images was proposed. The training data set was built with simulation data by the Monte Carlo method, and the self-developed TGS equipment was used for the experimental test. Experimental results show that ResNet can be used to reconstruct high-resolution TGS transmission images from sparse projection data. And compared with traditional algorithms, the proposed method has faster reconstruction speed and higher image quality.

Accurate and fast reconstruction for bioluminescence tomography based on adaptive Newton hard thresholding pursuit algorithm.

As a promising noninvasive medical imaging technique, bioluminescence tomography (BLT) dynamically offers three-dimensional visualization of tumor distribution in living animals. However, due to the high ill-posedness caused by the strong scattering property of biological tissues and the limited boundary measurements with noise, BLT reconstruction still cannot meet actual preliminary clinical application requirements. In our research, to recover 3D tumor distribution quickly and precisely, an adaptive Newton hard thresholding pursuit (ANHTP) algorithm is proposed to improve the performance of BLT. The ANHTP algorithm fully combines the advantages of sparsity constrained optimization and convex optimization to guarantee global convergence. More precisely, an adaptive sparsity adjustment strategy was developed to obtain the support set of the inverse system matrix. Based on the strong Wolfe line search criterion, a modified damped Newton algorithm was constructed to obtain optimal source distribution information. A series of numerical simulations and phantom and in vivo experiments show that ANHTP has high reconstruction accuracy, fast reconstruction speed, and good robustness. Our proposed algorithm can further increase the practicality of BLT in biomedical applications.

Iterator-Net: sinogram-based CT image reconstruction.

Image reconstruction is extremely important for computed tomography (CT) imaging, so it is significant to be continuously improved. The unfolding dynamics method combines a deep learning model with a traditional iterative algorithm. It is interpretable and has a fast reconstruction speed, but the essence of the algorithm is to replace the approximation operator in the optimization objective with a learning operator in the form of a convolutional neural network. In this paper, we firstly design a new iterator network (iNet), which is based on the universal approximation theorem and tries to simulate the functional relationship between the former and the latter in the maximum-likelihood expectation maximization (MLEM) algorithm. To evaluate the effectiveness of the method, we conduct experiments on a CT dataset, and the results show that our iNet method improves the quality of reconstructed images.

End-To-End Bioluminescence Tomography Reconstruction Based On Convolution Neural Network Scheme.

Bioluminescence tomography (BLT) has received a lot of attention as an important technique in bio-optical imaging. Compared with traditional methods, neural network methods have the advantages of fast reconstruction speed and support for batch processing. In this paper, we propose a end-to-end BLT reconstruction based on convolution neural networks scheme. First, 3000 datasets with single source and dual sources projection were conducted by Monte Carlo method, respectively. And three convolution neural networks (VGGNet, ResNet, and DenseNet) were adopted to feature extraction. Then, the filtered features were used as input to the multi-layer perceptron (MLP) to predict the source location. The results of numerical simulation and simulation experiments show, compared with traditional methods, the advantages of our method are including high reconstruction accuracy, faster reconstruction, few parameters, simple reconstruction process and support for batch processing.

Complementary time-frequency domain networks for dynamic parallel MR image reconstruction.

To introduce a novel deep learning-based approach for fast and high-quality dynamic multicoil MR reconstruction by learning a complementary time-frequency domain network that exploits spatiotemporal correlations simultaneously from complementary domains. Dynamic parallel MR image reconstruction is formulated as a multivariable minimization problem, where the data are regularized in combined temporal Fourier and spatial (x-f) domain as well as in spatiotemporal image (x-t) domain. An iterative algorithm based on variable splitting technique is derived, which alternates among signal de-aliasing steps in x-f and x-t spaces, a closed-form point-wise data consistency step and a weighted coupling step. The iterative model is embedded into a deep recurrent neural network which learns to recover the image via exploiting spatiotemporal redundancies in complementary domains. Experiments were performed on two datasets of highly undersampled multicoil short-axis cardiac cine MRI scans. Results demonstrate that our proposed method outperforms the current state-of-the-art approaches both quantitatively and qualitatively. The proposed model can also generalize well to data acquired from a different scanner and data with pathologies that were not seen in the training set. The work shows the benefit of reconstructing dynamic parallel MRI in complementary time-frequency domains with deep neural networks. The method can effectively and robustly reconstruct high-quality images from highly undersampled dynamic multicoil data ( and yielding 15 s and 10 s scan times respectively) with fast reconstruction speed (2.8 seconds). This could potentially facilitate achieving fast single-breath-hold clinical 2D cardiac cine imaging.

Light Weight IBP Deep Residual Network for Image Super Resolution

Single -image super resolution (SR) is used to reconstruct a high-resolution image with more high-frequency details based on a low-resolution image as input. In recent years, image SR reconstruction based on deep learning methods has shown a considerably better performance than traditional methods. Early deep-learning-based methods deepen convolutional layers and directly reconstruct high-resolution images with complex neural networks. However, with the stacking of modules, network depth and model parameters increase, thereby raising computational resource; hence, it is difficult to apply on low-configuration devices. Furthermore, existing methods ignore the high-frequency details of the image, resulting in unsatisfactory performance. To solve these problems, a lightweight network model that applies the iterative back projection (IBP) mechanism to the network and reduces the dimensionality of the input image features is proposed. The proposed network model consists of three parts, namely, entrance module, main body module, and exit module. It designs a lightweight modular design to reduce the model calculation and control the network depth more easily by adjusting the number of ADB modules. The main part of the model consists of four lightweight accelerating deep residual back projection (ADB) modules. Each ADB module initially decreases the dimensionality of the input image features through a 1×1 convolutional layer to reduce the amount of calculation. Then, IBP is used to back project the image iteratively. Each ADB module only performs the downsampling back projection operation because the image features become larger after upsampling. Through three iterative downsampling back projection units, the high-frequency features of the image are fully explored, and the output image features are then compared with the shallow layer. Image feature fusion is used as input of the next ADB module, and the output part combines the high- and low-frequency image feature output by multiple ADB modules to complete the upsampling by using the PixelShuffle method to generate high-resolution images. Several experiments confirm that the proposed algorithm achieves better SR reconstruction accuracy with faster reconstruction speed than existing image SR methods.

Using MCycleGAN to Realize High-Resolution Reconstruction of Solar Speckle Image

High-resolution reconstruction of solar speckle image is one of the important research contents in astronomical image processing. High-resolution image reconstruction based on deep learning can obtain the end-to-end mapping function from low-resolution image to high-resolution image through neural network model learning, which can recover the high-frequency information of the image. However, when used to reconstruct the sun speckle image with single feature, more noise and fuzzy local details, there are some shortcomings such as too smooth edge and easy loss of high-frequency information. In this paper, the structure features of input image and reconstructed image are added to CycleGAN network to get MCycleGAN. High frequency information is obtained from structural features by generator network, and the feature difference is calculated to enhance the ability of network to reconstruct high-frequency information. The edge of the reconstructed image is clearer. Compared with the speckle mask method level 1+ used by Yunnan Observatory, the results show that the proposed algorithm has the advantages of small error, fast reconstruction speed and high image clarity.

A feature-based convolutional neural network for reconstruction of interventional MRI.

Real-time interventional MRI (I-MRI) could help to visualize the position of the interventional feature, thus improving patient outcomes in MR-guided neurosurgery. In particular, in deep brain stimulation, real-time visualization of the intervention procedure using I-MRI could improve the accuracy of the electrode placement. However, the requirements of a high undersampling rate and fast reconstruction speed for real-time imaging pose a great challenge for reconstruction of the interventional images. Based on recent advances in deep learning (DL), we proposed a feature-based convolutional neural network (FbCNN) for reconstructing interventional images from golden-angle radially sampled data. The method was composed of two stages: (a) reconstruction of the interventional feature and (b) feature refinement and postprocessing. With only five radially sampled spokes, the interventional feature was reconstructed with a cascade CNN. The final interventional image was constructed with a refined feature and a fully sampled reference image. With a comparison of traditional reconstruction techniques and recent DL-based methods, it was shown that only FbCNN could reconstruct the interventional feature and the final interventional image. With a reconstruction time of ~ 500ms per frame and an acceleration factor of ~ 80, it was demonstrated that FbCNN had the potential for application in real-time I-MRI.

IFR-Net: Iterative Feature Refinement Network for Compressed Sensing MRI

To improve the compressive sensing MRI (CS-MRI) approaches in terms of fine structure loss under high acceleration factors, we have proposed an iterative feature refinement model (IFR-CS), equipped with fixed transforms, to restore the meaningful structures and details. Nevertheless, the proposed IFR-CS still has some limitations, such as the selection of hyper-parameters, a lengthy reconstruction time, and the fixed sparsifying transform. To alleviate these issues, we unroll the iterative feature refinement procedures in IFR-CS to a supervised model-driven network, dubbed IFR-Net. Equipped with training data pairs, both regularization parameter and the utmost feature refinement operator in IFR-CS become trainable. Additionally, inspired by the powerful representation capability of convolutional neural network (CNN), CNN-based inversion blocks are explored in the sparsity-promoting denoising module to generalize the sparsity-enforcing operator. Extensive experiments on both simulated and in vivo MR datasets have shown that the proposed network possesses a strong capability to capture image details and preserve well the structural information with fast reconstruction speed.

A randomised Kaczmarz method-based matrix completion algorithm for data collection in wireless sensor networks

The paper proposes a novel matrix completion algorithm for data collection in wireless sensor networks through incorporating a randomised version of Kaczmarz method. By splitting the matrix completion problem into two convex sub-problems and solving the optimal probability computing problem in the randomised Kaczmarz method approximately with the D-optimal design solution, we reduce the reconstruction error and accelerate the convergence speed of the matrix completion computation. The synthetic data experiments show that the proposed algorithm presents more accurate reconstruction accuracy and faster reconstruction speed than the state-of-the-art matrix completion algorithms. Furthermore, we verify the practicality of the proposed matrix completion algorithm in real data collection scenario of wireless sensor networks through the experiments based on the real sensed dataset.