Reconstruction Performance Research Articles

Supraharmonics Reconstruction Method Based on Blackman Window and Compressed Sensing

This paper proposes a method that combines window functions with compressed sensing for the detection of ultra-high harmonics in the frequency range of 2–150 kHz. By analyzing the sparsity of the signal, the Bruckmann window function, which is most suitable for the compressed sensing reconstruction process and the characteristics of ultra-high harmonics, is selected. Simulations indicate that, compared to existing methods, the proposed algorithm based on the fusion of the Bruckmann window and compressed sensing achieves a sparser post-observation signal with reduced fluctuations. The robustness and anti-interference capabilities are enhanced, while the harmonic detection accuracy and signal reconstruction performance are significantly improved. The reconstruction error reaches 4.15 × 10−6, the mean squared error percentage (MSE) reaches 4.13 × 10−6, and the signal-to-noise ratio (SNR) is as high as 97.69 dB, marking an increase of 54.11%. This study provides a new theoretical and methodological approach for the analysis and processing of ultra-high harmonics caused by a high proportion of power electronic devices.

Unwrapping non-locality in the image transmission through turbid media

Achieving high-fidelity image transmission through turbid media is a significant challenge facing both the AI and photonic/optical communities. While this capability holds promise for a variety of applications, including data transfer, neural endoscopy, and multi-mode optical fiber-based imaging, conventional deep learning methods struggle to capture the nuances of light propagation, leading to weak generalization and limited reconstruction performance. To address this limitation, we investigated the non-locality present in the reconstructed images and discovered that conventional deep learning methods rely on specific features extracted from the training dataset rather than meticulously reconstructing each pixel. This suggests that they fail to effectively capture long-range dependencies between pixels, which are crucial for accurate image reconstruction. Inspired by the physics of light propagation in turbid media, we developed a global attention mechanism to approach this problem from a broader perspective. Our network harnesses information redundancy generated by peculiar non-local features across the input and output fiber facets. This mechanism enables a two-order-of-magnitude performance boost and high fidelity to the data context, ensuring an accurate representation of intricate details in a pixel-to-pixel reconstruction rather than mere loss minimization.

DdeNet: A dual-domain end-to-end network combining Pale-Transformer and Laplacian convolution for sparse view CT reconstruction

Sparse view (SV) computed tomography (CT) is a clinical diagnostic technique aimed at reducing radiation dose to the human body from X-rays. However, SVCT reconstructed using conventional filtered back-projection (FBP) algorithms will produce severe streak artifacts. In recent years, convolutional neural networks (CNNs) have demonstrated considerable success in addressing the inverse problem of SVCT reconstruction. However, the local convolutional operations of CNNs ignores long-range dependencies of contextual feature information, thereby limiting their capability in feature extraction. To address this limitation, we propose a dual-domain end-to-end network (DdeNet) combining Pale-Transformer and Laplacian convolution for SVCT reconstruction. Initially, in the projection domain, a UNet is employed to restore the interpolated sinogram. Subsequently, in the image domain, we introduce Pale-Transformer and Laplacian convolution to design a dual-stream feature fusion network for finely restoring the CT images reconstructed by FBP. The dual-stream feature fusion network comprises an image restoration network (IRN) module and an edge enhancement network (EEN) module. The depthwise separable convolution based modified pale-shaped self-attention mechanism in Pale-Transformer captures global feature information of the CT image in the IRN, while the Laplacian convolution and multi-scale convolution block with a squeeze and excitation mechanism in the EEN further extract and restore edge feature information of the CT image. Finally, the feature information output from IRN and EEN is fused along the channel dimension and then convolved to obtain the final reconstructed CT image. Experimental results demonstrate that DdeNet exhibits superior generalization and reconstruction performance compared to the previous CNN or Transformer-based reconstruction methods.

Subgradient-projection-based stable phase-retrieval algorithm for X-ray ptychography.

X-ray ptychography is a lensless imaging technique that visualizes the nano-structure of a thick specimen which cannot be observed with an electron microscope. It reconstructs a complex-valued refractive index of the specimen from observed diffraction patterns. This reconstruction problem is called phase retrieval (PR). For further improvement in the imaging capability, including expansion of the depth of field, various PR algorithms have been proposed. Since a high-quality PR method is built upon a base PR algorithm such as ePIE, developing a well performing base PR algorithm is important. This paper proposes an improved iterative algorithm named CRISP. It exploits subgradient projection which allows adaptive step size and can be expected to avoid yielding a poor image. The proposed algorithm was compared with ePIE, which is a simple and fast-convergence algorithm, and its modified algorithm, rPIE. The experiments confirmed that the proposed method improved the reconstruction performance for both simulation and real data.

Reconstruction of missing wind data based on limited wind pressure measurements and machine learning

In structural health monitoring (SHM), wind field monitoring sometimes suffers from data loss owing to monitoring device failure, which inevitably creates barriers to subsequent data analysis and data mining. To this end, a novel strategy for reconstructing missing wind field data based on machine learning (ML) utilizing limited wind pressure measurements is proposed in this paper. Several ML algorithms, including decision tree, random forest, gradient boosting regression tree, support vector regression, Gaussian process regression, and backpropagation neural network, are employed to characterize potential relationships between wind pressure information (including time series and statistical parameters of wind pressures) and wind field information (e.g., wind direction and wind speed). Moreover, the effect of input information (including the type of input variables as well as the number and location of pressure transducers providing input data) on reconstruction performance and efficiency is investigated. Field measured records from an SHM system in a 600-m-high supertall building during typhoons are utilized to validate the feasibility and robustness of the proposed strategy. The results show that the presented strategy can effectively reconstruct missing wind field information in the SHM of the skyscraper during typhoons. Compared with the time series of wind pressures, selecting statistical parameters of wind pressures as input variables can effectively improve the performance and efficiency of reconstruction models. Choosing appropriate input information (e.g., using multiple input variables, adopting data from a larger number of pressure transducers, and utilizing data from pressure transducers closer to an anemometer) is beneficial for enhancing the performance of reconstruction models.

VNVC: A Versatile Neural Video Coding Framework for Efficient Human-Machine Vision.

Almost all digital videos are coded into compact representations before being transmitted. Such compact representations need to be decoded back to pixels before being displayed to humans and - as usual - before being enhanced/analyzed by machine vision algorithms. Intuitively, it is more efficient to enhance/analyze the coded representations directly without decoding them into pixels. Therefore, we propose a versatile neural video coding (VNVC) framework, which targets learning compact representations to support both reconstruction and direct enhancement/analysis, thereby being versatile for both human and machine vision. Our VNVC framework has a feature-based compression loop. In the loop, one frame is encoded into compact representations and decoded to an intermediate feature that is obtained before performing reconstruction. The intermediate feature can be used as reference in motion compensation and motion estimation through feature-based temporal context mining and cross-domain motion encoder-decoder to compress the following frames. The intermediate feature is directly fed into video reconstruction, video enhancement, and video analysis networks to evaluate its effectiveness. The evaluation shows that our framework with the intermediate feature achieves high compression efficiency for video reconstruction and satisfactory task performances with lower complexities.

Dual residual and large receptive field network for lightweight image super-resolution

In recent years, there has been rapid progress in single-image super-resolution (SISR) reconstruction technology based on deep learning, but many methods face challenges in practical application due to their excessive computational requirements. To address this issue, numerous lightweight super-resolution (SR) methods have been proposed. Convolutional neural networks (CNNs) have demonstrated outstanding performance in lightweight SR reconstruction tasks. However, existing CNN-based lightweight SR networks still suffer from redundancy due to the use of small convolutional kernels and deeper network architectures for feature extraction. To solve these problems, we propose a novel network called Dual Residual and Large Receptive Field Network (DRLN), which includes two main designs. Firstly, a Large Receptive Field Feature Extraction Block (LRFEB) is proposed, which concatenates large kernel depth separable convolutions and large kernel depth separable dilated convolutions, effectively capturing spatial contextual information without increasing the number of parameters or computational burden, and enhancing the overall performance of the model. Secondly, in the Efficient Feature Distillation Block (EFDB), we innovatively propose a dual residual structure, which greatly improve the learning ability and feature expression richness of the network, and significantly improve the reconstruction performance. Extensive experiments demonstrate that DRLN surpasses BSRN which is the champion of the model complexity track of the NTIRE 2022 Efficient SR Challenge on all benchmark datasets, with fewer parameters and less computation.

Spectral Super-Resolution via Model-Guided Cross-Fusion Network.

Spectral super-resolution, which reconstructs a hyperspectral image (HSI) from a single red-green-blue (RGB) image, has acquired more and more attention. Recently, convolution neural networks (CNNs) have achieved promising performance. However, they often fail to simultaneously exploit the imaging model of the spectral super-resolution and complex spatial and spectral characteristics of the HSI. To tackle the above problems, we build a novel cross fusion (CF)-based model-guided network (called SSRNet) for spectral super-resolution. In specific, based on the imaging model, we unfold the spectral super-resolution into the HSI prior learning (HPL) module and imaging model guiding (IMG) module. Instead of just modeling one kind of image prior, the HPL module is composed of two subnetworks with different structures, which can effectively learn the complex spatial and spectral priors of the HSI, respectively. Furthermore, a CF strategy is used to establish the connection between the two subnetworks, which further improves the learning performance of the CNN. The IMG module results in solving a strong convex optimization problem, which adaptively optimizes and merges the two features learned by the HPL module by exploiting the imaging model. The two modules are alternately connected to achieve optimal HSI reconstruction performance. Experiments on both the simulated and real data demonstrate that the proposed method can achieve superior spectral reconstruction results with relatively small model size. The code will be available at https://github.com/renweidian.

Enhanced hydrofoil pressure field reconstruction through phase-informed sensing and sensor optimization

This study proposes a cost-effective mixed-learning method for the cavitation behavior on hydrofoils, aiming to predict the global pressure field on the hydrofoil surface with sparse placed pressure sensors and the variation of the cavity outline captured by the high-speed cameras. The method integrates CS (compressed sensing) and deep learning techniques. Firstly, within the framework of CS, the Particle Swarm Optimization (PSO) algorithm is utilized to identify the best measuring points and achieve the initial reconstruction accordingly. Subsequently, in combination with the phase information, the final reconstructions are obtained within the deep learning framework. Through validation, the proposed mixed-learning model demonstrates significantly improved reconstruction performance compared to a single source of pressure information and an unoptimized measurement point position and significantly reduces the number of sensors. This provides a novel and effective approach for accurately predicting the pressure field on hydrofoil surfaces. Furthermore, the study evaluates the robustness concerning hydraulic sensor measurement errors, sensor placement deviations, missing measuring points, reconstruction performance under various cases and hydrofoil surface regions, and the contribution of optimized measuring points to the global field. Results show that the mixed-learning model exhibits robustness against typical measurement errors, position deviations of hydraulic sensors and missing measuring points on the suction side. Additionally, a positive correlation exists between the average pressure gradient under different conditions and regions and the reconstruction error, with the pressure side region significantly contributing to global field reconstruction. These findings underscore the mixed model's superiority in predicting pressure fields on hydrofoils and offer guidance for rational sensor installation.

CTIS spectral image reconstruction technology based on slit-scanning architecture

The computed tomography imaging spectrometer (CTIS) is a snapshot imaging spectrometer, excelling in dynamic detection tasks. It can capture two-dimensional spatial information and spectrally compressed information of a target within a single exposure time. However, traditional CTIS image reconstruction algorithms suffer from missing-cone problem, which reduces the accuracy of spectral reconstruction. In recent years, deep learning has been applied to CTIS spectral image reconstruction, significantly improving spectral reconstruction accuracy compared to traditional algorithms. However, due to the missing-cone problem, it is difficult to accurately recover the truth of spectral data cube in the real scene. Currently, most CTIS neural network reconstruction models are trained using simulated datasets of spectral data cubes and diffractive images. Because these data differ significantly from real data under actual application conditions, the established models may not be effectively applicable to real-world scenes. Therefore, we propose a new CTIS system based on slit-scanning architecture utilizing an adjustable slit aperture to obtain the real spectral data cube of the target while maintaining the simplicity of the CTIS structure. By limiting the field of view (FOV) through the slit, the area of diffraction overlap can be reduced, thereby enhancing the accuracy of CTIS spectral reconstruction using the expectation-maximization (EM) algorithm. This architecture allows us to obtain accurate spectral cubes that match the CTIS diffractive image of real-world scenes, providing a real dataset for training the reconstruction network. A prototype has been built to demonstrate the feasibility of our proposed solution. Furthermore, we also constructed a residual network based on multi-scale and attention mechanism. This network is trained using a combination of simulated and real spectral imaging data. Compared to the reconstruction performance of the EM algorithm and convolutional neural networks, our approach demonstrates superior spectral reconstruction accuracy, validating the importance of real spectral data in CTIS spectral reconstruction tasks.

Visually secure traffic image encryption scheme using new two-dimensional Sigmoid-type memristive chaotic map and Laguerre transform embedding

In intelligent transportation system, unprotected bare data transmission faces serious security threats and challenges. To this end, this paper proposes a visually secure traffic image encryption scheme that combines a newly designed two-dimensional Sigmoid-type memristive chaotic map (2D-SMCM) with two-dimensional compressive sensing (2D-CS) and Laguerre transform (LT) embedding to provide services for secure transmission of private images. Specifically, first, the 2D-SMCM is used to generate pseudo-random sequences for subsequent compression, encryption and hiding operations. Second, the 2D-CS is utilized to compress the plain image to reduce the amount of data transmission. Then, encryption is completed by modifying the data values and their positions through index permutation and bidirectional diffusion. Finally, the encrypted data is embedded in the LT-processed public carrier medium for covert transmission. Experiments and performance analysis illustrate that the proposed scheme has good security, imperceptibility and reconstruction performance, with the average PSNRs of the cipher images and decrypted secret images up to 45.90 dB and 34.85 dB, respectively, using 500 grayscale images from the database BOWS2.

SRBPSwin: Single-Image Super-Resolution for Remote Sensing Images Using a Global Residual Multi-Attention Hybrid Back-Projection Network Based on the Swin Transformer

Remote sensing images usually contain abundant targets and complex information distributions. Consequently, networks are required to model both global and local information in the super-resolution (SR) reconstruction of remote sensing images. The existing SR reconstruction algorithms generally focus on only local or global features, neglecting effective feedback for reconstruction errors. Therefore, a Global Residual Multi-attention Fusion Back-projection Network (SRBPSwin) is introduced by combining the back-projection mechanism with the Swin Transformer. We incorporate a concatenated Channel and Spatial Attention Block (CSAB) into the Swin Transformer Block (STB) to design a Multi-attention Hybrid Swin Transformer Block (MAHSTB). SRBPSwin develops dense back-projection units to provide bidirectional feedback for reconstruction errors, enhancing the network’s feature extraction capabilities and improving reconstruction performance. SRBPSwin consists of the following four main stages: shallow feature extraction, shallow feature refinement, dense back projection, and image reconstruction. Firstly, for the input low-resolution (LR) image, shallow features are extracted and refined through the shallow feature extraction and shallow feature refinement stages. Secondly, multiple up-projection and down-projection units are designed to alternately process features between high-resolution (HR) and LR spaces, obtaining more accurate and detailed feature representations. Finally, global residual connections are utilized to transfer shallow features during the image reconstruction stage. We propose a perceptual loss function based on the Swin Transformer to enhance the detail of the reconstructed image. Extensive experiments demonstrate the significant reconstruction advantages of SRBPSwin in quantitative evaluation and visual quality.

A ridge-based detection algorithm with filament overlap identification for 2D mycelium network analysis

Many terrestrial ecosystems engage mycorrhizal symbiotic associations, potentially to enhance nutrition, increase resistance to soil-borne pests and diseases, and improve resilience and soil structure. Mycorrhizal fungi create dynamic networked structures through branching and anastomosis that connect multiple plants and consent to transport resources underground from nutrient-rich patches to demanding plants. Controlled laboratory experiments are fundamental to improving our knowledge of mycelium network growth dynamics and further understanding its role in preserving ecological niches. We propose a method for highly automated analysis of the mycelium network structure and other morphological properties, such as hyphal length, hyphal density, and number of crossing and branches, in 2D microscopy images of fungal samples. Available tools for automated network analyses suffer from overestimating network connectivity since filament crossings are not considered. In particular, we propose a) a ridge-based mycelium detection algorithm and b) a geometrical-based approach to identify overlapping filaments crossing each other. The algorithmic solution is evaluated on a total of 135 real mycelium sample images over different validation steps, originating from different datasets and having different characteristics, including background, contrast, image acquisition system, fungal species, and clearness (e.g., level of transparency, homogeneity, dirtiness of the medium) of the sample. Results show that 1) the proposed detection method can be used to measure the length of mycelium in an image, replacing manual tracing and allowing for less laborious analysis ρ̂c=0.96, 2) the filament detection is on par with state-of-the-art techniques F1=0.88−0.94 with a more intuitive parameterization, and 3) the proposed algorithm correctly identifies filament crossings F1=0.89 in most common cases, yielding a reduction in the overestimation of network connectivity. The latter feature consents to applying the proposed fully automated solution to complex and irregular fungal structures, advancing mycelium detection and reconstruction performance accuracy with respect to the state-of-the-art.

High-quality AFM image acquisition of living cells by modified residual encoder-decoder network

Atomic force microscope enables ultra-precision imaging of living cells. However, atomic force microscope imaging is a complex and time-consuming process. The obtained images of living cells usually have low resolution and are easily influenced by noise leading to unsatisfactory imaging quality, obstructing the research and analysis based on cell images. Herein, an adaptive attention image reconstruction network based on residual encoder-decoder was proposed, through the combination of deep learning technology and atomic force microscope imaging supporting high-quality cell image acquisition. Compared with other learning-based methods, the proposed network showed higher peak signal-to-noise ratio, higher structural similarity and better image reconstruction performances. In addition, the cell images reconstructed by each method were used for cell recognition, and the cell images reconstructed by the proposed network had the highest cell recognition rate. The proposed network has brought insights into the atomic force microscope-based imaging of living cells and cell image reconstruction, which is of great significance in biological and medical research.

An interpretable and flexible fusion prior to boost hyperspectral imaging reconstruction

Hyperspectral image (HSI) reconstruction from the compressed measurement captured by the coded aperture snapshot spectral imager system remains a hot topic. Recently, deep-learning-based methods for HSI reconstruction have become the mainstream due to their high performance and efficiency in the testing inference. However, these learning methods do not fully utilize the abundant spectral information with proper physical spectral priors, resulting in complex architectures and unsatisfactory reconstruction performance. In this paper, we claim that the spectral low-rank property can still help these learning methods and propose a hyperspectral fusion theory, which demonstrates that full HSIs are mathematically equivalent to the closed-form combination of subspace images, mask, and measurements. Based on the above fusion theory, we propose the subspace distillation prior (SP) to efficiently cooperate with existing learning models to enhance the exploration of the spectral low-rank property. In detail, the SP can directly improve the testing inference of existing models (SP1, Section 4.1). Furthermore, SP can also be cooperated with exiting networks to formulate a new framework, which regularizes the existing models to learn the subspace images, and help to reconstruct the full HSIs from subspace images, mask, and measurements (SP2, Section 4.2). We choose six existing representative models for the HSI reconstruction experiments and find that SP1 and SP2 can, respectively, achieve improvements of 0.08 dB∼0.76 dB and 0.36 dB∼1.76 dB on the simulated datasets, demonstrating the advantage of the proposed hyperspectral fusion theory. The source code is available at https://github.com/prowDIY.

Super-resolution reconstruction of ultrasound image using a modified diffusion model

Objective. This study aims to perform super-resolution (SR) reconstruction of ultrasound images using a modified diffusion model, designated as the diffusion model for ultrasound image super-resolution (DMUISR). SR involves converting low-resolution images to high-resolution ones, and the proposed model is designed to enhance the suitability of diffusion models for this task in the context of ultrasound imaging. Approach. DMUISR incorporates a multi-layer self-attention (MLSA) mechanism and a wavelet-transform based low-resolution image (WTLR) encoder to enhance its suitability for ultrasound image SR tasks. The model takes interpolated and magnified images as input and outputs high-quality, detailed SR images. The study utilized 1,334 ultrasound images from the public fetal head-circumference dataset (HC18) for evaluation. Main results. Experiments were conducted at 2× , 4× , and 8× magnification factors. DMUISR outperformed mainstream ultrasound SR methods (Bicubic, VDSR, DECUSR, DRCN, REDNet, SRGAN) across all scales, providing high-quality images with clear structures and rich detailed textures in both hard and soft tissue regions. DMUISR successfully accomplished multiscale SR reconstruction while suppressing over-smoothing and mode collapse problems. Quantitative results showed that DMUISR achieved the best performance in terms of learned perceptual image patch similarity, with a significant decrease of over 50% at all three magnification factors (2× , 4× , and 8× ), as well as improvements in peak signal-to-noise ratio and structural similarity index measure. Ablation experiments validated the effectiveness of the MLSA mechanism and WTLR encoder in improving DMUISR’s SR performance. Furthermore, by reducing the number of diffusion steps, the computational time of DMUISR was shortened to nearly one-tenth of its original while maintaining image quality without significant degradation. Significance. This study demonstrates that the modified diffusion model, DMUISR, provides superior performance for SR reconstruction of ultrasound images and has potential in improving imaging quality in the medical ultrasound field.

A Low-Bit-Rate Image Semantic Communication System Based on Semantic Graph

In the progress of research in the field of semantic communication, most efforts have been focused on optimizing the signal-to-noise ratio (SNR), while the design and optimization of the bit rate required for transmission have been relatively neglected. To address this issue, this study introduces an innovative low-bit-rate image semantic communication system model, which aims to reconstruct images through the exchange of semantic information rather than traditional symbol transmission. This model employs an image feature extraction and optimization reconstruction framework, achieving visually satisfactory and semantically consistent reconstruction performance at extremely low bit rates (below 0.03 bits per pixel (bpp)). Unlike previous methods that used pixel accuracy as the standard for distortion measurement, this research introduces multiple perceptual metrics to train and evaluate the proposed image semantic encoding model, aligning more closely with the fundamental purpose of semantic communication. Experimental results demonstrate that, compared to WebP, JPEG, and deep learning-based image codecs, our model not only provides a more visually pleasing reconstruction effect but also significantly reduces the required bit rate while maintaining semantic consistency.

Analysis of medical images super-resolution via a wavelet pyramid recursive neural network constrained by wavelet energy entropy

Recently, multi-resolution pyramid-based techniques have emerged as the prevailing research approach for image super-resolution. However, these methods typically rely on a single mode of information transmission between levels. In our approach, a wavelet pyramid recursive neural network (WPRNN) based on wavelet energy entropy (WEE) constraint is proposed. This network transmits previous-level wavelet coefficients and additional shallow coefficient features to capture local details. Besides, the parameter of low- and high-frequency wavelet coefficients within each pyramid level and across pyramid levels is shared. A multi-resolution wavelet pyramid fusion (WPF) module is devised to facilitate information transfer across network pyramid levels. Additionally, a wavelet energy entropy loss is proposed to constrain the reconstruction of wavelet coefficients from the perspective of signal energy distribution. Finally, our method achieves the competitive reconstruction performance with the minimal parameters through an extensive series of experiments conducted on publicly available datasets, which demonstrates its practical utility.

Response reconstruction based on substructural condensation and modal-group superposition

Abstract Increasingly complex civil structures result in an enormous consumption of computational resources in response reconstruction. Moreover, common closely spaced modes (CSMs) also impair reconstruction performance. This study proposed an improved approach that incorporated substructural condensation and modal-group superposition. This method efficiently extrapolated the desired response by dividing the entire finite element model into multiple manageable substructures, which were then condensed and assembled into a super-element model with fewer degrees of freedom (DOFs). The reconstruction response was directly performed on the condensed model by selectively retaining the physically measured DOFs during the condensation process. To address the challenges posed by CSMs, defined modal group responses (MGRs) were extracted from dynamic measurements using empirical mode decomposition. Based on these results, the corresponding MGRs at inaccessible locations were extrapolated using a condensed model. Finally, all the MGRs are superposed to form the desired response. Numerical and experimental tests were conducted to demonstrate the efficiency and accuracy of the proposed approach. The effects of substructural division, measurement noise, and measurement locations on the reconstruction were studied in detail.

Recovering high-quality fiber orientation distributions from a reduced number of diffusion-weighted images using a model-driven deep learning architecture.

The aim of this study was to develop a model-based deep learning architecture to accurately reconstruct fiber orientation distributions (FODs) from a reduced number of diffusion-weighted images (DWIs), facilitating accurate analysis with reduced acquisition times. Our proposed architecture, Spherical Deconvolution Network (SDNet), performed FOD reconstruction by mapping 30 DWIs to fully sampled FODs, which have been fit to 288 DWIs. SDNet included DWI-consistency blocks within the network architecture, and a fixel-classification penalty within the loss function. SDNet was trained on a subset of the Human Connectome Project, and its performance compared with FOD-Net, and multishell multitissue constrained spherical deconvolution. SDNet achieved the strongest results with respect to angular correlation coefficient and sum of squared errors. When the impact of the fixel-classification penalty was increased, we observed an improvement in performance metrics reliant on segmenting the FODs into the correct number of fixels. Inclusion of DWI-consistency blocks improved reconstruction performance, and the fixel-classification penalty term offered increased control over the angular separation of fixels in the reconstructed FODs.