Denoising Network Research Articles

Image Denoising Using Deblur Generative Adversarial Network Denoising U-Net

Convolutional neural networks (CNNs) are becoming increasingly popular for image denoising. U-Nets, a type of CNN architecture, have been shown to be effective for this task. However, the impact of shallow layers on deeper layers decreases as the depth of the network increases. To address this issue, the authors propose a new image denoising method called DGANDU-Net. DGANDU-Net combines the DeblurGAN design with a U-Net architecture. This combination allows DGANDU-Net to effectively remove noise from images while preserving fine details. The authors also propose the use of two loss functions, mean square error (MSE) and perceptual loss, to improve the performance of DGANDU-Net. MSE is used to learn and improve the extracted features, while perceptual loss is used to produce the final denoised image. The authors evaluate the performance of DGANDU-Net on a variety of noise levels and find that it outperforms other state-of-the-art denoising algorithms in terms of both visual quality and two evaluation indices, including peak signal-to-noise ratio (PSNR) and Structural Similarity Index Measure (SSIM). Specifically, for extremely noisy environments with a noise standard deviation of 75, DGANDU-Net achieves an average PSNR of 37.39[Formula: see text]dB in the test dataset. The authors conclude that DGANDU-Net is a promising new method for image denoising that has the potential to significantly improve the quality of medical images used for diagnosis and treatment.

Noise-assisted hybrid attention networks for low-dose PET and CT denoising.

Positron emission tomography (PET) and computed tomography (CT) play a vital role in tumor-related medical diagnosis, assessment, and treatment planning. However, full-dose PET and CT pose the risk of excessive radiation exposure to patients, whereas low-dose images compromise image quality, impacting subsequent tumor recognition and diseasediagnosis. To solve such problems, we propose a Noise-Assisted Hybrid Attention Network (NAHANet) to reconstruct full-dose PET and CT images from low-dose PET (LDPET) and CT (LDCT) images to reduce patient radiation risks while ensuring the performance of subsequent tumor recognition. NAHANet contains two branches: the noise feature prediction branch (NFPB) and the cascaded reconstruction branch. Among them, NFPB providing noise features for the cascade reconstruction branch. The cascaded reconstruction branch comprises a shallow feature extraction module and a reconstruction module which contains a series of cascaded noise feature fusion blocks (NFFBs). Among these, the NFFB fuses the features extracted from low-dose images with the noise features obtained by NFPB to improve the feature extraction capability. To validate the effectiveness of the NAHANet method, we performed experiments using two public available datasets: the Ultra-low Dose PET Imaging Challenge dataset and Low Dose CT Grand Challengedataset. As a result, the proposed NAHANet achieved higher performance on common indicators. For example, on the CT dataset, the PSNR and SSIM indicators were improved by 4.1dB and 0.06 respectively, and the rMSE indicator was reduced by 5.46 compared with the LDCT; on the PET dataset, the PSNR and SSIM was improved by 3.37dB and 0.02, and the rMSE was reduced by 9.04 compared with theLDPET. This paper proposes a transformer-based denoising algorithm, which utilizes hybrid attention to extract high-level features of low dose images and fuses noise features to optimize the denoising performance of the network, achieving good performance improvements on low-dose CT and PETdatasets.

MFCA-MICNN: a convolutional neural network with multiscale fast channel attention and multibranch irregular convolution for noise removal in dMRI

Diffusion magnetic resonance imaging (dMRI) currently stands as the foremost noninvasive method for quantifying brain tissue microstructure and reconstructing white matter fiber pathways. However, the inherent free diffusion motion of water molecules in dMRI results in signal decay, diminishing the signal-to-noise ratio (SNR) and adversely affecting the accuracy and precision of microstructural data. In response to this challenge, we propose a novel method known as the Multiscale Fast Attention-Multibranch Irregular Convolutional Neural Network for dMRI image denoising. In this work, we introduce Multiscale Fast Channel Attention, a novel approach for efficient multiscale feature extraction with attention weight computation across feature channels. This enhances the model's capability to capture complex features and improves overall performance. Furthermore, we propose a multi-branch irregular convolutional architecture that effectively disrupts spatial noise correlation and captures noise features, thereby further enhancing the denoising performance of the model. Lastly, we design a novel loss function, which ensures excellent performance in both edge and flat regions. Experimental results demonstrate that the proposed method outperforms other state-of-the-art deep learning denoising methods in both quantitative and qualitative aspects for dMRI image denoising with fewer parameters and faster operational speed.

Memorizing Swin-Transformer Denoising Network for Diffusion Model

Diffusion models have garnered significant attention in the field of image generation. However, existing denoising architectures, such as U-Net, face limitations in capturing the global context, while Vision Transformers (ViTs) may struggle with local receptive fields. To address these challenges, we propose a novel Swin-Transformer-based denoising network architecture that leverages the strengths of both U-Net and ViT. Moreover, our approach integrates the k-Nearest Neighbor (kNN) based memorizing attention module into the Swin-Transformer, enabling it to effectively harness crucial contextual information from feature maps and enhance its representational capacity. Finally, we introduce an innovative hierarchical time stream embedding scheme that optimizes the incorporation of temporal cues during the denoising process. This method surpasses basic approaches like simple addition or concatenation of fixed time embeddings, facilitating a more effective fusion of temporal information. Extensive experiments conducted on four benchmark datasets demonstrate the superior performance of our proposed model compared to U-Net and ViT as denoising networks. Our model outperforms baselines on the CRC-VAL-HE-7K and CelebA datasets, achieving improved FID scores of 14.39 and 4.96, respectively, and even surpassing DiT and UViT under our experiment setting. The Memorizing Swin-Transformer architecture, coupled with the hierarchical time stream embedding, sets a new state-of-the-art in denoising diffusion models for image generation.

De novo protein design with a denoising diffusion network independent of pretrained structure prediction models.

The recent success of RFdiffusion, a method for protein structure design with a denoising diffusion probabilistic model, has relied on fine-tuning the RoseTTAFold structure prediction network for protein backbone denoising. Here, we introduce SCUBA-diffusion (SCUBA-D), a protein backbone denoising diffusion probabilistic model freshly trained by considering co-diffusion of sequence representation to enhance model regularization and adversarial losses to minimize data-out-of-distribution errors. While matching the performance of the pretrained RoseTTAFold-based RFdiffusion in generating experimentally realizable protein structures, SCUBA-D readily generates protein structures with not-yet-observed overall folds that are different from those predictable with RoseTTAFold. The accuracy of SCUBA-D was confirmed by the X-ray structures of 16 designed proteins and a protein complex, and by experiments validating designed heme-binding proteins and Ras-binding proteins. Our work shows that deep generative models of images or texts can be fruitfully extended to complex physical objects like protein structures by addressing outstanding issues such as the data-out-of-distribution errors.

SITF: A Self-Supervised Iterative Training Framework for Point Cloud Denoising

Despite existing supervised point cloud denoising methods having made great progress, they require paired ideal noisy-clean datasets for training which is expensive and impractical in real-world applications. Moreover, they may perform the denoising process multiple times with fixed network parameters for better denoising results at test time. To address above issues, this paper proposes a self-supervised iterative training framework (SITF) for point cloud denoising, which only requires single noisy point clouds and a noise model. Given an off-the-shelf denoising network and original noisy point clouds, firstly, an intermediate noisier-noisy dataset is created by adding additional noises from the known noise model to noisy point clouds (i.e. learning targets). Secondly, after training on the noisier-noisy dataset, the denoising network is employed to denoise the original noisy point clouds to obtain the learning targets for the next iteration. The above two steps are iteratively and alternatively performed to get a better and better trained denoising network. Furthermore, to get better learning targets for the next round, this paper also proposes a novel iterative denoising network (IDN) architecture of stacked source attention denoising modules. The IDN explicitly models the iterative denoising process internally within a single network via reforming the given denoising network. Experimental results show that existing supervised networks trained through the SITF can achieve competitive denoising results and even outperform supervised networks under high noise conditions. The source code can be found at: https://github.com/VCG-NJUST/SITF.

Metric learning guided sinogram denoising for cone beam CT enhancement.

Cone beam computed tomography (CBCT) is a widely available modality, but its clinical utility has been limited by low detail conspicuity and quantitative accuracy. Convenient post-reconstruction denoising is subject to back projected patterned residual, but joint denoise-reconstruction is typically computationally expensive and complex. In this study, we develop and evaluate a novel Metric-learning guided wavelet transform reconstruction (MEGATRON) approach to enhance image domain quality with projection-domain processing. Projection domain based processing has the benefit of being simple, efficient, and compatible with various reconstruction toolkit and vendor platforms. However, they also typically show inferior performance in the final reconstructed image, because the denoising goals in projection and image domains do not necessarily align. Motivated by these observations, this work aims to translate the demand for quality enhancement from the quantitative image domain to the more easily operable projection domain. Specifically, the proposed paradigm consists of a metric learning module and a denoising network module. Via metric learning, enhancement objectives on the wavelet encoded sinogram domain data are defined to reflect post-reconstruction image discrepancy. The denoising network maps measured cone-beam projection to its enhanced version, driven by the learnt objective. In doing so, the denoiser operates in the convenient sinogram to sinogram fashion but reflects improvement in reconstructed image as the final goal. Implementation-wise, metric learning was formalized as optimizing the weighted fitting of wavelet subbands, and a res-Unet, which is a Unet structure with residual blocks, was used for denoising. To access quantitative reference, cone-beam projections were simulated using the X-ray based Cancer Imaging Simulation Toolkit (XCIST). In both learning modules, a data set of 123 human thoraxes, which was from Open-Source Imaging Consortium (OSIC) Pulmonary Fibrosis Progression challenge, was used. Reconstructed CBCT thoracic images were compared against ground truth FB and performance was assessed in root mean square error (RMSE), peak signal-to-noise ratio (PSNR), and structural similarity index (SSIM). MEGATRON achieved RMSE in HU value, PSNR, and SSIM were 30.97±4.25, 37.45±1.78, and 93.23±1.62, respectively. These values are on par with reported results from sophisticated physics-driven CBCT enhancement, demonstrating promise and utility of the proposed MEGATRON method. We have demonstrated that incorporating the proposed metric learning into sinogram denoising introduces awareness of reconstruction goal and improves final quantitative performance. The proposed approach is compatible with a wide range of denoiser network structures and reconstruction modules, to suit customized need or further improve performance.

Unsupervised Adaptation Learning for Real Multiplatform Hyperspectral Image Denoising.

Real hyperspectral images (HSIs) are ineluctably contaminated by diverse types of noise, which severely limits the image usability. Recently, transfer learning has been introduced in hyperspectral denoising networks to improve model generalizability. However, the current frameworks often rely on image priors and struggle to retain the fidelity of background information. In this article, an unsupervised adaptation learning (UAL)-based hyperspectral denoising network (UALHDN) is proposed to address these issues. The core idea is first learning a general image prior for most HSIs, and then adapting it to a real HSI by learning the deep priors and maintaining background consistency, without introducing hand-crafted priors. Following this notion, a spatial-spectral residual denoiser, a global modeling discriminator, and a hyperspectral discrete representation learning scheme are introduced in the UALHDN framework, and are employed across two learning stages. First, the denoiser and the discriminator are pretrained using synthetic noisy-clean ground-based HSI pairs. Subsequently, the denoiser is further fine-tuned on the real multiplatform HSI according to a spatial-spectral consistency constraint and a background consistency loss in an unsupervised manner. A hyperspectral discrete representation learning scheme is also designed in the fine-tuning stage to extract semantic features and estimate noise-free components, exploring the deep priors specific for real HSIs. The applicability and generalizability of the proposed UALHDN framework were verified through the experiments on real HSIs from various platforms and sensors, including unmanned aerial vehicle-borne, airborne, spaceborne, and Martian datasets. The UAL denoising scheme shows a superior denoising ability when compared with the state-of-the-art hyperspectral denoisers.

HPIDN: A Hierarchical prior-guided iterative denoising network with global–local fusion for enhancing low-dose CT images

Low-dose computed tomography (LDCT) is an emerging medical diagnostic tool that reduces radiation exposure but suffers from noise retention. Current CNN-based LDCT denoising algorithms struggle to capture comprehensive global representations, impacting diagnostic accuracy. To address this, we propose a novel Hierarchical Prior-guided Iterative Denoising Network (HPIDN) for LDCT images, consisting of two main modules: the Dynamic Feature Extraction and Fusion Module (DFEFM) and the Feature-domain Iterative Denoising Module (FIDM). DFEFM dynamically captures a comprehensive representation, encompassing detailed local features in intra-relationships and complex global features in inter-relationships. It effectively guides the multi-stage iterative denoising process. FIDM hierarchically fuses the prior with image features from DFEFM by using the dual-domain attention fusion sub-network (DAFSN), enhancing denoising robustness and adaptability. This yields higher-quality images with reduced noise artifacts. Extensive experiments on the Mayo and ELCAP Datasets demonstrate the superior performance of our method quantitatively and qualitatively, improving diagnostic accuracy of lung diseases.

ScVGATAE: A Variational Graph Attentional Autoencoder Model for Clustering Single-Cell RNA-seq Data.

Single-cell RNA sequencing (scRNA-seq) is now a successful technology for identifying cell heterogeneity, revealing new cell subpopulations, and predicting developmental trajectories. A crucial component in scRNA-seq is the precise identification of cell subsets. Although many unsupervised clustering methods have been developed for clustering cell subpopulations, the performance of these methods is prone to be affected by dropout, high dimensionality, and technical noise. Additionally, most existing methods are time-consuming and fail to fully consider the potential correlations between cells. In this paper, we propose a novel unsupervised clustering method called scVGATAE (Single-cell Variational Graph Attention Autoencoder) for scRNA-seq data. This method constructs a reliable cell graph through network denoising, utilizes a novel variational graph autoencoder model integrated with graph attention networks to aggregate neighbor information and learn the distribution of the low-dimensional representations of cells, and adaptively determines the model training iterations for various datasets. Finally, the obtained low-dimensional representations of cells are clustered using kmeans. Experiments on nine public datasets show that scVGATAE outperforms classical and state-of-the-art clustering methods.

A Novel Network for Low-Dose CT Denoising Based on Dual-Branch Structure and Multi-Scale Residual Attention.

Deep learning-based denoising of low-dose medical CT images has received great attention both from academic researchers and physicians in recent years, and has shown important application value in clinical practice. In this work, a novel two-branch and multi-scale residual attention-based network for low-dose CT image denoising is proposed. It adopts a two-branch framework structure, to extract and fuse image features at shallow and deep levels respectively, to recover image texture and structure information as much as possible. We propose the adaptive dynamic convolution block (ADCB) in the local information extraction layer. It can effectively extract the detailed information of low-dose CT denoising and enables the network to better capture the local details and texture features of the image, thereby improving the denoising effect and image quality. Multi-scale edge enhancement attention block (MEAB) is proposed in the global information extraction layer, to perform feature fusion through dilated convolution and a multi-dimensional attention mechanism. A multi-scale residual convolution block (MRCB) is proposed to integrate feature information and improve the robustness and generalization of the network. To demonstrate the effectiveness of our method, extensive comparison experiments are conducted and the performances evaluated on two publicly available datasets. Our model achieves 29.3004 PSNR, 0.8659 SSIM, and 14.0284 RMSE on the AAPM-Mayo dataset. It is evaluated by adding four different noise levels σ = 15, 30, 45, and 60 on the Qin_LUNG_CT dataset and achieves the best results. Ablation studies show that the proposed ADCB, MEAB, and MRCB modules improve the denoising performances significantly. The source code is available at https://github.com/Ye111-cmd/LDMANet .

Adaptive edge prior-based deep attention residual network for low-dose CT image denoising

Improving the diagnostic quality of low-dose CT (LDCT) images relies on effective noise removal. Recent advancements have highlighted the widespread use of deep residual networks for LDCT image denoising. These networks possess properties that aid in preserving image integrity and optimizing model performance. However, the denoising process faces challenges due to the complex patterns and intensity similarities between edge details and lesion regions. To address this issue, this paper introduces a novel approach called the cross-scale attentional residual network (RCANet), which utilizes an adaptive edge prior for LDCT image denoising. The adaptive edge prior enhances the denoising network’s ability to retain image boundary features and uniqueness. To distinguish subtle differences between LDCT image edge details and lesion areas, a cross-scale mapping dual-element module (CMDM) is designed to preserve rich edge texture information during model training. To prevent over-smoothing of denoised results, a compound loss function is proposed, combining MSE loss and multi-scale attention residual perception loss. To validate the effectiveness of the method, experiments were conducted on the AAPM-Mayo Clinic LDCT Grand Challenge dataset. The results demonstrate that RCANet surpasses state-of-the-art residual structure-based network models and performs comparably to leading denoising algorithms.

Convolutional neural network transformer (CNNT) for fluorescence microscopy image denoising with improved generalization and fast adaptation

Deep neural networks can improve the quality of fluorescence microscopy images. Previous methods, based on Convolutional Neural Networks (CNNs), require time-consuming training of individual models for each experiment, impairing their applicability and generalization. In this study, we propose a novel imaging-transformer based model, Convolutional Neural Network Transformer (CNNT), that outperforms CNN based networks for image denoising. We train a general CNNT based backbone model from pairwise high-low Signal-to-Noise Ratio (SNR) image volumes, gathered from a single type of fluorescence microscope, an instant Structured Illumination Microscope. Fast adaptation to new microscopes is achieved by fine-tuning the backbone on only 5–10 image volume pairs per new experiment. Results show that the CNNT backbone and fine-tuning scheme significantly reduces training time and improves image quality, outperforming models trained using only CNNs such as 3D-RCAN and Noise2Fast. We show three examples of efficacy of this approach in wide-field, two-photon, and confocal fluorescence microscopy.

Self-Supervised Image Denoising of Third Harmonic Generation Microscopic Images of Human Glioma Tissue by Transformer-Based Blind Spot (TBS) Network.

Third harmonic generation (THG) microscopy shows great potential for instant pathology of brain tumor tissue during surgery. However, due to the maximal permitted exposure of laser intensity and inherent noise of the imaging system, the noise level of THG images is relatively high, which affects subsequent feature extraction analysis. Denoising THG images is challenging for modern deep-learning based methods because of the rich morphologies contained and the difficulty in obtaining the noise-free counterparts. To address this, in this work, we propose an unsupervised deep-learning network for denoising of THG images which combines a self-supervised blind spot method and a U-shape Transformer using a dynamic sparse attention mechanism. The experimental results on THG images of human glioma tissue show that our approach exhibits superior denoising performance qualitatively and quantitatively compared with previous methods. Our model achieves an improvement of 2.47-9.50 dB in SNR and 0.37-7.40 dB in CNR, compared to six recent state-of-the-art unsupervised learning models including Neighbor2Neighbor, Blind2Unblind, Self2Self+, ZS-N2N, Noise2Info and SDAP. To achieve an objective evaluation of our model, we also validate our model on public datasets including natural and microscopic images, and our model shows a better denoising performance than several recent unsupervised models such as Neighbor2Neighbor, Blind2Unblind and ZS-N2N. In addition, our model is nearly instant in denoising a THG image, which has the potential for real-time applications of THG microscopy.

Speckle noise suppression of a reconstructed image in digital holography based on the BM3D improved convolutional neural network

In digital holographic measurement, when light waves pass through inhomogeneous media or surfaces, speckle noise is generated, resulting in random, granular light and dark spots in the hologram, which greatly reduces the image quality. Therefore, in order to improve the image quality of holographic reconstruction, a noise reduction method based on the BM3D improved convolutional neural network (CNN) is proposed in this paper. Firstly, the similarity and important statistical information between blocks can be obtained by using BM3D. Then, the denoising convolutional neural network (DnCNN) is used to learn the relationship between the noise of a large number of samples and the noise image, and further purify the image to retain the details for a better denoising effect. Finally, a reflective off-axis digital holographic optical path system is constructed to collect the holograms of the test samples, and the reconstructed images are obtained by the Fresnel diffraction method to constitute a dataset with the simulated holographic reconstructed images to validate the proposed method in this paper, compared to the other methods, such as DnCNN, convolutional blind denoising network (CBDNet), BM3D, and Wiener filtering. The experimental results of qualitative and quantitative analyses show that the proposed method combines the advantages of traditional algorithms and deep learning, significantly enhances the robustness of the system, optimizes the denoising performance, and preserves the details of the reconstructed image to the greatest extent.

Enhanced Learning Enriched Features Mechanism Using Deep Convolutional Neural Network for Image Denoising and Super-Resolution

Enhanced Learning Enriched Features Mechanism Using Deep Convolutional Neural Network for Image Denoising and Super-Resolution

Anti‐interference lithium‐ion battery intelligent perception for thermal fault detection and localization

AbstractLithium‐ion batteries are widely employed in electric vehicles, power grid energy storage, and other fields. Thermal fault diagnostics for battery packs is crucial to preventing thermal runaway from impairing the safe operation and extended cycle service life of batteries. Therefore, a lithium‐ion battery thermal fault diagnosis model based on deep learning algorithms is presented, which includes three parts: autoencoder denoising network, coarse mask generator, and mask precise adjustment. Autoencoder denoising network can reduce data noise during thermal imaging acquisition, improve the anti‐interference ability of diagnostic models, and ensure the accuracy of thermal runaway diagnosis. A two‐stage diagnostic structure is then formulated by the coarse mask generator and mask precise adjustment, which enable quick identification, categorisation, and localisation of thermal fault battery cells. According to the test results, the segmentation boundary is more distinct and is capable of matching the original image's level. The recognition accuracy of the thermal diagnosis model for faulty batteries is close to 100%. After denoising by the autoencoder, the prediction results improved by 22% compared to non‐local mean denoising and by about 32% compared to noisy images.

Road Target Detection in Different Weather Conditions Based on Deep Learning

AbstractAddressing the challenge of ensuring robustness in vision‐based target recognition algorithms under adverse weather conditions, such as rain, snow, and fog, is crucial. In this paper, we introduce a novel approach for road target detection that can effectively operate under various weather conditions. Our method is based on the cascade task framework of target detection, complemented by image restoration techniques. Specifically, we have developed a denoising network tailored to meet the demands of de‐raining and snow removal tasks. This network leverages prior knowledge about the mask, enhancing its effectiveness. In real‐world scenarios featuring fog, wet conditions, and snow‐covered roads, our proposed method demonstrates a significant improvement in both recall rate and accuracy compared to conventional single‐object detection algorithms. © 2024 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.

Wavelet and Adaptive Coordinate Attention Guided Fine-Grained Residual Network for Image Denoising

Wavelet and Adaptive Coordinate Attention Guided Fine-Grained Residual Network for Image Denoising

A Residual U-Net Neural Network for Seismocardiogram Denoising and Analysis During Physical Activity.

Seismocardiogram (SCG) signals are noninvasively obtained cardiomechanical signals containing important features for cardiovascular health monitoring. However, these signals are prone to contamination by motion noise, which can significantly impact accuracy and robustness of the measurements. A deep learning model based on the U-Net architecture is proposed to recover SCG signals contaminated by motion noise induced by walking. The model performance was evaluated through qualitative visualization, as well as quantitative analyses. Quantitative analyses included distance-based comparisons before and after applying our model. Analyses also included assessments of the model's efficacy in improving the performance of downstream tasks related to health parameter estimation during walking. Experimental findings revealed that the denoising model improved similarity to clean signals by approximately 90%. The performance of the model in enhancing heart rate estimation demonstrated a mean absolute error of 1.21 BPM and a root-mean-squared error (RMSE) of 1.97 BPM during walking after denoising with 9.16 BPM and 10.38 BPM improvements, respectively, compared to without denoising. Furthermore, the RMSEs of aortic opening and aortic closing time estimation after denoising for one dataset with catheter ground truth were 7.29 ms and 19.71 ms during walking, respectively, with 50.33 ms and 51.91 ms RMSE improvements compared to without denoising. And for another dataset with ICG-derived PEP ground truth, the RMSE of aortic opening time estimation after denoising was 10.21 ms during walking, with 38.74 ms RMSE improvement compared to without denoising. The proposed model attenuates motion noise from corrupted SCG signals while preserving cardiac information. This development paves the way for improved ambulatory cardiac health monitoring using wearable accelerometers during daily activities.