High Reconstruction Accuracy Research Articles

Three-Dimensional Subject-Specific Knee Shape Reconstruction with Asynchronous Fluoroscopy Images Using Statistical Shape Modeling.

Background and objectives: Statistical shape modeling (SSM) based on computerized tomography (CT) datasets has enabled reasonably accurate reconstructions of subject-specific 3D bone morphology from one or two synchronous radiographs for clinical applications. Increasing the number of radiographic images may increase the reconstruction accuracy, but errors related to the temporal and spatial asynchronization of clinical alternating bi-plane fluoroscopy may also increase. The current study aimed to develop a new approach for subject-specific 3D knee shape reconstruction from multiple asynchronous fluoroscopy images from 2, 4, and 6 X-ray detector views using a CT-based SSM model; and to determine the optimum number of planar images for best accuracy via computer simulations and in vivo experiments. Methods: A CT-based SSM model of the knee was established from 60 training models in a healthy young Chinese male population. A new two-phase optimization approach for 3D subject-specific model reconstruction from multiple asynchronous clinical fluoroscopy images using the SSM was developed, and its performance was evaluated via computer simulation and in vivo experiments using one, two and three image pairs from an alternating bi-plane fluoroscope. Results: The computer simulation showed that subject-specific 3D shape reconstruction using three image pairs had the best accuracy with RMSE of 0.52 ± 0.09 and 0.63 ± 0.085 mm for the femur and tibia, respectively. The corresponding values for the in vivo study were 0.64 ± 0.084 and 0.69 ± 0.069 mm, respectively, which was significantly better than those using one image pair (0.81 ± 0.126 and 0.83 ± 0.108 mm). No significant differences existed between using two and three image pairs. Conclusion: A new two-phase optimization approach was developed for SSM-based 3D subject-specific knee model reconstructions using more than one asynchronous fluoroscopy image pair from widely available alternating bi-plane fluoroscopy systems in clinical settings. A CT-based SSM model of the knee was also developed for a healthy young Chinese male population. The new approach was found to have high mode reconstruction accuracy, and those for both two and three image pairs were much better than for a single image pair. Thus, two image pairs may be used when considering computational costs and radiation dosage. The new approach will be useful for generating patient-specific knee models for clinical applications using multiple asynchronous images from alternating bi-plane fluoroscopy widely available in clinical settings. The current SSM model will serve as a basis for further inclusion of training models with a wider range of sizes and morphological features for broader applications.

Synthetic coded aperture snapshot spectral imaging based on coprime sub-aperture sampling.

The imaging of a large area scene is difficult to achieve for a single camera. Alternatively, a virtual large aperture can be synthesized by sub-aperture cooperation. We propose a solution through the combination of a coded aperture snapshot spectral imager and coprime array. This method reduces the amount of data by using a smaller sub-aperture for sampling. The position of the sub-aperture is determined according to the element distribution of the coprime array, so that the data obtained from each sampling are about the target object and its adjacent area, which can ensure high reconstruction accuracy. The feasibility of coprime sub-aperture sampling is verified by numerical simulation.

A fast low rank Vandermonde factorization reconstruction method for non-uniformly sampled 2D NMR spectroscopy

As an indispensable analytical method in the fields of chemistry and medicine, nuclear magnetic resonance (NMR) spectroscopy is widely adopted for analyzing the composition and structure of substances. However, the data acquisition period for high-resolution NMR spectra is extremely long. Therefore, non-uniform sampling is commonly adopted to reduce the measurement time and the complete NMR signal can be obtained by appropriate reconstruction algorithms. Recently, it was found that NMR signal can be reconstructed well by low-rank Hankel matrix completion. However, the computational time required for this method to reconstruct two-dimensional (2D) and higher dimensional signals is relatively long because it increases rapidly as spectral resolution improves. In this paper, we propose a novel low rank minimization method to alleviate this problem by utilizing the unexploited property that a 2D NMR signal can be decomposed into two Vandermonde matrices. However, minimizing the rank necessitates the time-consuming singular value decomposition (SVD). Thus we introduce a matrix bi-factorization idea into the method to reduce the time complexity and greatly mitigate the time required by the SVD. Experiments on synthetic data and realistic NMR spectroscopy data demonstrate that compared with the state-of-the-art nuclear norm based minimization and fast iterative hard thresholding algorithms, the computational time of the proposed algorithm is significantly reduced, and high reconstruction accuracy can be ensured at the same time.

Neural Network Approximation of Graph Fourier Transform for Sparse Sampling of Networked Dynamics

Infrastructure monitoring is critical for safe operations and sustainability. Like many networked systems, water distribution networks (WDNs) exhibit both graph topological structure and complex embedded flow dynamics. The resulting networked cascade dynamics are difficult to predict without extensive sensor data. However, ubiquitous sensor monitoring in underground situations is expensive, and a key challenge is to infer the contaminant dynamics from partial sparse monitoring data. Existing approaches use multi-objective optimization to find the minimum set of essential monitoring points but lack performance guarantees and a theoretical framework. Here, we first develop a novel Graph Fourier Transform (GFT) operator to compress networked contamination dynamics to identify the essential principal data collection points with inference performance guarantees. As such, the GFT approach provides the theoretical sampling bound. We then achieve under-sampling performance by building auto-encoder (AE) neural networks (NN) to generalize the GFT sampling process and under-sample further from the initial sampling set, allowing a very small set of data points to largely reconstruct the contamination dynamics over real and artificial WDNs. Various sources of the contamination are tested, and we obtain high accuracy reconstruction using around 5%–10% of the network nodes for known contaminant sources, and 50%–75% for unknown source cases, which although larger than that of the schemes for contaminant detection and source identifications, is smaller than the current sampling schemes for contaminant data recovery. This general approach of compression and under-sampled recovery via NN can be applied to a wide range of networked infrastructures to enable efficient data sampling for digital twins.

Can the Structure Similarity of Training Patches Affect the Sea Surface Temperature Deep Learning Super-Resolution?

Meso- and fine-scale sea surface temperature (SST) is an essential parameter in oceanographic research. Remote sensing is an efficient way to acquire global SST. However, single infrared-based and microwave-based satellite-derived SST cannot obtain complete coverage and high-resolution SST simultaneously. Deep learning super-resolution (SR) techniques have exhibited the ability to enhance spatial resolution, offering the potential to reconstruct the details of SST fields. Current SR research focuses mainly on improving the structure of the SR model instead of training dataset selection. Different from generating the low-resolution images by downscaling the corresponding high-resolution images, the high- and low-resolution SST are derived from different sensors. Hence, the structure similarity of training patches may affect the SR model training and, consequently, the SST reconstruction. In this study, we first discuss the influence of training dataset selection on SST SR performance, showing that the training dataset determined by the structure similarity index (SSIM) of 0.6 can result in higher reconstruction accuracy and better image quality. In addition, in the practical stage, the spatial similarity between the low-resolution input and the objective high-resolution output is a key factor for SST SR. Moreover, the training dataset obtained from the actual AMSR2 and MODIS SST images is more suitable for SST SR because of the skin and sub-skin temperature difference. Finally, the SST reconstruction accuracies obtained from different SR models are relatively consistent, yet the differences in reconstructed image quality are rather significant.

Structured sparsity learning for large-scale fuzzy cognitive maps

Fuzzy cognitive map (FCM) as a kind of intelligent soft computing method, by combining the advantages of neural network and fuzzy logic, can be used to mine the causal relationships between concepts and make reasoning. However, how to effectively learn the large-scale FCMs is still an open problem. In this article, by means of structured sparsity learning, a robust learning method for large-scale FCMs based on iterative smoothing algorithm is proposed. Firstly, in terms of sparse signal reconstruction, the objective function of learning method is constructed by using elastic and total variation (TV) penalties, which can be conducive to capture the sparse structure information of FCM and improve the robustness of network reconstruction. Due to the non-smoothness of the TV penalty, Nesterov’s smoothing technique is used to solve the non-smooth problem, thus transforming the problem into a convex optimization problem. Subsequently, in order to quickly solve the convex optimization, the algorithm based on proximal gradient descent is applied. In the experiment part, synthetic FCM models with different densities, sizes and noises are used to evaluate the proposed method, and the experimental results demonstrate the proposed method can make full use of the observations to learn the structural information of FCM. Moreover, the real-world data from the gene regulatory networks (GRNs) are further used to evaluate the effect of network reconstruction, and a higher reconstruction accuracy can be verified.

Network Reconstruction in Terms of the Priori Structure Information

In this paper, we investigate the reconstruction of networks based on priori structure information by the Element Elimination Method (EEM). We firstly generate four types of synthetic networks as small-world networks, random networks, regular networks and Apollonian networks. Then, we randomly delete a fraction of links in the original networks. Finally, we employ EEM, the resource allocation (RA) and the structural perturbation method (SPM) to reconstruct four types of synthetic networks with 90% priori structure information. The experimental results show that, comparing with RA and SPM, EEM has higher indices of reconstruction accuracy on four types of synthetic networks. We also compare the reconstruction performance of EEM with RA and SPM on four empirical networks. Higher reconstruction accuracy, measured by local indices of success rates, could be achieved by EEM, which are improved by 64.11 and 47.81%, respectively.

Homotopic Gradients of Generative Density Priors for MR Image Reconstruction.

Deep learning, particularly the generative model, has demonstrated tremendous potential to significantly speed up image reconstruction with reduced measurements recently. Rather than the existing generative models that often optimize the density priors, in this work, by taking advantage of the denoising score matching, homotopic gradients of generative density priors (HGGDP) are exploited for magnetic resonance imaging (MRI) reconstruction. More precisely, to tackle the low-dimensional manifold and low data density region issues in generative density prior, we estimate the target gradients in higher-dimensional space. We train a more powerful noise conditional score network by forming high-dimensional tensor as the network input at the training phase. More artificial noise is also injected in the embedding space. At the reconstruction stage, a homotopy method is employed to pursue the density prior, such as to boost the reconstruction performance. Experiment results implied the remarkable performance of HGGDP in terms of high reconstruction accuracy. Only 10% of the k-space data can still generate image of high quality as effectively as standard MRI reconstructions with the fully sampled data.

Accurate 3D reconstruction of complex defects based on combined method of MFL and MFDs

The two-step strategy for 3D reconstruction of complex defects was updated by combining the magnetic field disturbance (MFD) with magnetic flux leakage (MFL) measurement methods. Firstly, the plane-scanned MFD imaging was employed to achieve high-accuracy reconstruction of the opening outlines of complex defects. Secondly, the finite element simulation model of the defect with the recognized opening outline was established as the forward model for depth profile inversion with the particle swarm optimization algorithm. The performances of the conventional MFL-based and updated strategies were carefully compared through finite element simulation. The results proved that the updated method combining MFD signals with MFL signals had the better performance in accurate 3D reconstruction of complex defects than the conventional MFL-based inversion strategy, displaying smaller errors in opening outline, peak depth and the profile. Experimental results confirmed that the updated strategy could accurately reconstruct the 3D shape of 8-shaped defect and rectangular defect with rounded corners.

Predicting optical transmission through complex scattering media from reflection patterns with deep neural networks

Deep neural networks (DNNs) are used to reconstruct transmission speckle intensity patterns from the respective reflection speckle intensity patterns generated by illuminated parafilm layers. The dependence of the reconstruction accuracy on the thickness of the sample is examined for different illumination patterns of various feature sizes. High reconstruction accuracy is obtained even for large parafilm thicknesses, for which the memory effect of the sample is vanishingly small. The generalization capability of the DNN is also studied for unseen scatterers of the same type.

Reconstruction Method for Missing Measurement Data Based on Wasserstein Generative Adversarial Network

Focusing on the issue of missing measurement data caused by complex and changeable working conditions during the operation of high-speed trains, in this paper, a framework for the reconstruction of missing measurement data based on a generative adversarial network is proposed. Suitable parameters were set for each frame. Discrete measurement data are taken as the input of the frame for preprocessing the data dimensionality. The convolutional neural network then learns the correlation between different characteristic values of each device in an unsupervised pattern and constrains and improves the reconstruction accuracy by taking advantage of the context similarity of authenticity. It was determined experimentally that when there are different extents of missing measurement data, the model described in the present paper can still maintain a high reconstruction accuracy. In addition, the reconstruction data also conform well to the distribution law of the measurement data.

Multiparametric Electromagnetic Inversion of 3-D Anisotropic Objects Embedded in Layered Media Based on Mixed $L_1$–$L_2$ Norm Regularization

In this letter, the mixed $L_1$ – $L_2$ norm regularization is adopted to constrain the full-wave inversion solution of the multiparametric electromagnetic inverse scattering by the 3-D anisotropic objects. The mixed $L_1$ – $L_2$ regularization term is added to the standard $L_2$ norm cost function in the framework of the Born iterative method. In each iteration, the mixed $L_1$ – $L_2$ norm term is transformed into $L_2$ norm term through a weight matrix and the whole cost function is minimized by the conjugate gradient method. Numerical results show that, compared with the traditional Tikhonov $L_2$ regularization, the mixed $L_1$ – $L_2$ regularization can achieve higher reconstruction accuracy without additional computational cost due to the structural similarity constraint among different model parameters.

Learning-Based Sparse Data Reconstruction for Compressed Data Aggregation in IoT Networks

Due to the booming of various devices in Internet-of-Things (IoT) networks, more data should be transmitted over the networks, which will thereby consume more transmission bandwidth and more transmit power. Compressed data aggregation (CDA) has been proposed as an effective way to reduce the amount of the collected data in IoT networks. Although adopting compressed sensing (CS), CDA can sample the source data efficiently, and sparse data reconstruction is still a big challenge. In this work, inspired by this, we propose a learning-based sparse data reconstruction scheme by jointly utilizing CS and deep learning. Our objective is to reduce the volume of data to be transmitted over IoT networks without losing reconstruction accuracy. A deep CS network is designed by adopting an end-to-end learning method to build a measurement matrix and an efficient and high-accuracy reconstruction network. To show the performance of the proposed scheme, six data sets with different structured sparse models and a real sensor data set are utilized in doing experiments. The performance of the proposed scheme in terms of mean-squared error, peak-signal-noise-ratio, and structural similarity is investigated. The results demonstrate the effectiveness of the proposed scheme in reconstruction accuracy for given compression ratio. The results also show that the proposed scheme is suitable for the process of CDA, thus can effectively reduce the amount of data to be transmitted in IoT networks.

M2H-Net: A Reconstruction Method For Hyperspectral Remotely Sensed Imagery

Hyperspectral remote sensing can get spatially and spectrally continuous data simultaneously. However, the imaging equipment is usually expensive and complex, along with the low spatial resolution. In recent years, reconstruction of hyperspectral image by deep learning from the widely used low-cost, high spatial resolution RGB camera, has attracted extensive attention in many fields. However, most research is limited to three bands in the range of 400–700 nm, which greatly restrains its application in remote sensing. In this study, a more suitable for remote sensing multispectral to hyperspectral network (M2H-Net) is proposed, which can take many bands as input and output hyperspectral images with any number of bands within a wider spectral range (380–2500 nm). Its characteristics include adding residual connection on U-Net to reduce vanishing gradients; adding convolution combinations with different kernel sizes (1 × 1 and 3 × 3) to balance the spectral and spatial relationships. It is applied on images from different platforms (UAVs and Satellites), different imaging modes (frame and pushbroom) and different spectral response functions (narrow and wide bandwidth), and the results show that: 1) it has a very high accuracy of hyperspectral image reconstruction. The mean relative absolute error (MRAE) and root mean squared error (RMSE) are between 0.039 and 0.074 and 0.010–0.016, respectively, which are 69.2% and 41.2% lower than those of U-Net; 2) it has high efficiency with fast convergence (about 40 epochs) and stable performance. Compared with many algorithms won in the new trends in image restoration and enhancement (NTIRE) competition, M2H-Net ranked 7th in accuracy, but took less time (0.44 s); 3) it has strong generalization ability. Using the pre-trained M2H-Nets to reconstruct Cubert S185 and GF-5 hyperspectral images in different locations, different times and complex scenes, high accuracy (MRAE = 0.072, RMSE = 0.011) can still be obtained. This method is more suitable for remote sensing to meet the needs of multiple bands, spectrum width and complex scenes, thus provides the possibility to generate the global coverage hyperspectral imagery by using the massive in-orbit or historical archived multispectral images, which will not only greatly save the R&D and investment on hyperspectral imaging equipment, but also conduct data collection with higher efficiency and lower complexity. Due to the ability to reconstruct hyperspectral images in specified bands on demand, M2H-Net is also of great value in hyperspectral image processing, such as data compression, storage and transmission, etc.

A Task-Driven Invertible Projection Matrix Learning Algorithm for Hyperspectral Compressed Sensing

The high complexity of the reconstruction algorithm is the main bottleneck of the hyperspectral image (HSI) compression technology based on compressed sensing. Compressed sensing technology is an important tool for retrieving the maximum number of HSI scenes on the ground. However, the complexity of the compressed sensing algorithm is limited by the energy and hardware of spaceborne equipment. Aiming at the high complexity of compressed sensing reconstruction algorithm and low reconstruction accuracy, an equivalent model of the invertible transformation is theoretically derived by us in the paper, which can convert the complex invertible projection training model into the coupled dictionary training model. Besides, aiming at the invertible projection training model, the most competitive task-driven invertible projection matrix learning algorithm (TIPML) is proposed. In TIPML, we don’t need to directly train the complex invertible projection model, but indirectly train the invertible projection model through the training of the coupled dictionary. In order to improve the accuracy of reconstructed data, in the paper, the singular value transformation is proposed. It has been verified that the concentration of the dictionary is increased and that the expressive ability of the dictionary has not been reduced by the transformation. Besides, two-loop iterative training is established to improve the accuracy of data reconstruction. Experiments show that, compared with the traditional compressed sensing algorithm, the compressed sensing algorithm based on TIPML has higher reconstruction accuracy, and the reconstruction time is shortened by more than a hundred times. It is foreseeable that the TIPML algorithm will have a huge application prospect in the field of HSI compression.

Calibration Procedure of Camera With Multifocus Zoom Lens for Three-Dimensional Scanning System

With zoom lenses, both the magnification of the camera and the distance between the lens and the object plane can be changed. Changes to zoom lens magnification is accompanied by changes in the depth of field and the distance between the lens and the object plane, complicating the calibration of zoom cameras-specifically, that of each focus and zoom control, for which separate configurations are required. To overcome this problem, a procedure for the calibration of a multifocus zoom lens in a three-dimensional (3D) scanning system was developed in the present study. Specifically, a pragmatic, five-step calibration procedure was proposed. In the first two steps, the characteristics of the zoom lens were obtained. In the third and fourth steps, the calibration was normalized. Finally, the effectiveness of the process was assessed using a laboratory-developed prototype. High reconstruction accuracy was achieved.

Latent Variable Models Reconstruct Diversity of Neuronal Response to Drifting Gratings in Murine Visual Cortex

Decades of experimentation in visual neuroscience have been conducted with limited sample sizes of simultaneously recorded neurons. Past experiments have characterized the visual system based on statistical properties of individual neurons. Yet, no neuron functions in isolation: neural networks code information through their coordinated behavior. Calcium imaging can now simultaneously record larger populations than electrophysiological techniques could previously, allowing single-neuron analyses to be expanded to the network level. Larger datasets demand the development of computational approaches for network-level analysis. Our work validates a population modelling approach, utilizing two latent variable models and a large dataset of simultaneously recorded neurons. Rectified Latent Variable Model (RLVM) and Non-Negative Matrix Factorization (NNMF) were able to be fit to a large calcium imaging dataset without deconvolving the fluorescence signal into action potentials. These models show high reconstruction accuracy, with the RLVM outperforming NNMF. Both models’ reconstruction performance can be predicted via regression with node density and cell behavior metrics as predictors, showing that RLVM and NNMF are more effective on networks with high node density. Neurons and estimated latent variables show a spectrum of functional groups in response to simple stimulus features, with groups ranging from orientation and frequency specific, or orientation specific frequency agnostic, to orientation specific frequency modulated. Clustering of behavior profiles reveals some similar response profile groups between neurons and estimated latent variables in response to drifting gratings stimuli.

Incoherent digital holographic spectral imaging with high accuracy of image pixel registration* *Project supported by the National Natural Science Foundation of China (Grant Nos. 61505178, 61307019, and 11504333) and the Natural Science Foundation of Henan Province, China (Grant Nos. 18A140032, 15A140038, and 16A140035).

Fresnel incoherent correlation holography (FINCH) is a unique three-dimensional (3D) imaging technique which has the advantages of scanning-free, high resolution, and easy matching with existing mature optical systems. In this article, an incoherent digital holographic spectral imaging method with high accuracy of spectral reconstruction based on liquid crystal tunable filter (LCTF) and FINCH is proposed. Using the programmable characteristics of spatial light modulator (SLM), a series of phase masks, none of whose focal lengths changes with wavelength, is designed and made. For each wavelength of LCTF output, SLM calls three phase masks with different phase constants at the corresponding wavelength, and CCD records three holograms. The spectral images obtained by this method have a constant magnification, which can achieve pixel-level image registration, restrain image registration errors, and improve spectral reconstruction accuracy. The results show that this method can not only obtain the 3D spatial information and spectral information of the object simultaneously, but also have high accuracy of spectral reconstruction and excellent color reproducibility.

Modal decomposition methods for distributed excitation force field on tube bundle in cross flow

To rapidly and accurately identify the characteristics of the distributed force and reconstruct the force field on tube bundles in a cross flow, modal decomposition methods for a three-dimensional turbulent flow are studied. This is done with proper orthogonal decomposition (POD), dynamic mode decomposition (DMD), and the regularization method for DMD. A 3-D turbulent cross flow computational fluid dynamics model is simulated to obtain the temporal–spatial force field acting on the tubes. Applying direct Fourier transformation to the data yields the frequency contents of the field as a reference for other analyses. POD provides high-accuracy reconstruction of the force field with 43 modes, and each mode contains multiple frequency contents of the Fourier transformation. In comparison, DMD successfully demonstrates the relation between single frequencies and the spatial distribution on the tube surface. To solve the ill-posedness of singular value decomposition (SVD), the key mathematical processing of both decomposition methods, the Tikhonov regularization method is introduced into field reconstruction. The regularization operator and parameter determined by solving a functional extremum problem yield a criterion that can be applied to the SVD of the DMD, which effectively reduces the number of DMD modes required to reconstruct the field to 16 with the same level of norm error.

Generative adversarial networks for dual-modality electrical tomography in multi-phase flow measurement

In many multi-phase flows, the online measurement and monitoring of phase fractions as well as distributions play a vital role in determining the process efficiency and safety. The dual-modality electrical capacitance tomography (ECT) and electromagnetic tomography (EMT) technique provides an efficient measure to estimate the distribution of electromagnetic property in rapidly changing multi-phase flows. A generative adversarial network (GAN) is designed to solve the fusion problem of ECT and EMT in the gas–liquid–solid (G–L-S) three-phase flow measurement. The fusion model incorporates the features of dual-modality measurements and images to generate the electromagnetic property in high precision, by which the accurate phase volume can be derived. Furthermore, a simulation approach is proposed to provide the sufficient measurement samples that approximate the real three-phase flow measurement. In the numerical study, the fluidization process of a G–L–S fluidized bed (GLSFB) reactor is simulated and measured by the models of ECT and EMT. The simulation validation on samples from GLSFB and experiments on the three-phase flow setup demonstrate the high accuracy of electromagnetic property reconstruction and generalization ability of fusion model that suitable for various flow regimes. The errors of calculated phase fraction are less than 0.15 in both simulations and experiments.