Self-supervised neural networks have demonstrated remarkable performance in image-denoising applications. However, existing dataset-free methods have limitations, including their high computational overhead, noise mod...
详细信息
Self-supervised neural networks have demonstrated remarkable performance in image-denoising applications. However, existing dataset-free methods have limitations, including their high computational overhead, noise model requirements, and compromised image quality. To address these problems, we proposed a dataset-free method for image denoising using a single noisy image. The proposed method uses a lightweight neural network with slightly more than 3000 parameters to achieve effective denoising. Inspired by the Zero-Shot Noise2Noise framework, we downsampled noisy images and employed downscaled nonfused images for training, achieving denoising by cross-mapping the denoised subimages with the original noisy subimages. To further reduce the high-frequency noise, we employed cross-mapping among the denoised subimages. The numerical results demonstrated the superior performance of the proposed algorithm compared with other dataset-free neural network algorithms. The proposed method exhibited a shorter processing time and fewer network parameters yet yielded denoised images with a higher signal-to-noise ratio. Moreover, it demonstrated superior performance and processing speed in both synthetic and actual noise experiments, making it suitable for practical applications.
For solving linear inverse problems, particularly of the type that appears in tomographic imaging and compressive sensing, this paper develops two new approaches. The first approach is an iterative algorithm that mini...
详细信息
For solving linear inverse problems, particularly of the type that appears in tomographic imaging and compressive sensing, this paper develops two new approaches. The first approach is an iterative algorithm that minimizes a regularized least squares objective function where the regularization is based on a compound Gaussian prior distribution. The compound Gaussian prior subsumes many of the commonly used priors in image reconstruction, including those of sparsity-based approaches. The developed iterative algorithm gives rise to the paper's second new approach, which is a deep neural network that corresponds to an "unrolling" or "unfolding" of the iterative algorithm. Unrolled deep neural networks have interpretable layers and outperform standard deep learning methods. This paper includes a detailed computational theory that provides insight into the construction and performance of both algorithms. The conclusion is that both algorithms outperform other state-of-the-art approaches to tomographic image formation and compressive sensing, especially in the difficult regime of low training.
To achieve a visually captivating nocturnal image that closely resembles its natural daytime counterpart, people employ a range of techniques to process the nighttime image. The primary focus lies in achieving rapid a...
详细信息
To achieve a visually captivating nocturnal image that closely resembles its natural daytime counterpart, people employ a range of techniques to process the nighttime image. The primary focus lies in achieving rapid and stable unsupervised image enhancement effects specifically tailored for nocturnal scenes, without relying on daytime contrast image. However, existing neural network-based methods for enhancing nighttime image often rely on supervised paired training data, which presents challenges in practical production scenarios. The acquisition of image pairs depicting the same scene and the creation of a large-scale, feature-rich training dataset pose significant difficulties. In this study, we propose a fast pure nighttime image enhancement technique based on preprocessing inspired by the varying light sensitivity exhibited by fish during night fishing. The sensitivity of fish to light varies at different depths, analogous to the concealed richness of effective information within seemingly dark nighttime image, which can be effectively and comprehensively unveiled through preprocessing techniques. Subsequently, we employ an improved dual logarithmic imageprocessing method based on type-II fuzzy sets to fuse the layer information obtained from preprocessing, resulting in enhanced contrast, noise reduction, color enhancement, and improved illumination with superior quality. The extensive experimental and comparative results demonstrate that our method's robust enhancement and restoration capabilities surpass even those of state-of-the-art supervised methods.
Spike camera is a bio-inspired sensor with ultra-high temporal resolution and low energy consumption. It captures visual signals using an "integrate-and-fire" mechanism and outputs a continuous stream of bin...
详细信息
Spike camera is a bio-inspired sensor with ultra-high temporal resolution and low energy consumption. It captures visual signals using an "integrate-and-fire" mechanism and outputs a continuous stream of binary spikes. Reconstructing image sequence from spikes streams is critical for spike camera. Several reconstruction methods have been proposed in recent years. However, the computational cost of these methods is relatively high. Inspired by the fact that spiking neural networks (SNNs) are energy efficient and support time-series signalprocessing inherently, we propose a lightweight SNN for spike camera image reconstruction (abbreviated to SSIR). Experimental results show that SSIR achieves comparable performance with the state-of-the-art (SOTA) methods at much lower computation and energy cost.
Gastric cancer is a leading cause of cancer-related deaths globally. As mortality rates continue to rise, predicting cancer survival using multimodal data-including histopathological images, genomic data, and clinical...
详细信息
Gastric cancer is a leading cause of cancer-related deaths globally. As mortality rates continue to rise, predicting cancer survival using multimodal data-including histopathological images, genomic data, and clinical information-has become increasingly crucial. However, extracting effective predictive features from this complex data has posed challenges for survival analysis due to the high dimensionality and heterogeneity of histopathology images and genomic data. Furthermore, existing methods often lack sufficient interaction between intra- and inter- modal features, significantly impacting model performance. To address these challenges, we developed a deep learning-based multimodal feature fusion model, MultiDeepsurv, designed to predict the survival of gastric cancer patients by integrating histopathological images, clinical data, and gene expression data. Our approach includes a two-branch hybrid network, GLFUnet, which leverages the attention mechanism for enhanced pathology image representation learning. Additionally, we employ a graph convolutional neural network (GCN) to extract features from gene expression data and clinical information. To capture the correlations between different modalities, we utilize the SFusion fusion strategy that employs a self-attention mechanism to learn potential correlations across modalities. Finally, these deeply processed features are fed into Cox regression models for an end-to-end survival analysis. Comprehensive experiments and analyses conducted on a gastric cancer cohort from The Cancer Genome Atlas (TCGA) demonstrate that our proposed MultiDeepsurv model outperforms other methods in terms of prognostic accuracy, with a C-index of 0.806 and an AUC of 0.842. (c) 2024 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
Purpose: Proton magnetic resonance spectroscopic imaging (1H-MRSI) provides non-invasive spectral-spatial mapping of metabolism. However, long-standing problems in whole-brain 1H-MRSI are spectral overlap of metabolit...
详细信息
Purpose: Proton magnetic resonance spectroscopic imaging (1H-MRSI) provides non-invasive spectral-spatial mapping of metabolism. However, long-standing problems in whole-brain 1H-MRSI are spectral overlap of metabolite peaks with large lipid signal from scalp, and overwhelming water signal that distorts spectra. Fast and effective methods are needed for high-resolution 1H-MRSI to accurately remove lipid and water signals while preserving the metabolite signal. The potential of supervised neural networks for this task remains unexplored, despite their success for other MRSI processing. methods: We introduce a deep learning method based on a modified Y-NET network for water and lipid removal in whole-brain 1H-MRSI. The WALINET (WAter and LIpid neural NETwork) was compared with conventional methods such as the state-of-the-art lipid L2 regularization and Hankel-Lanczos singular value decomposition (HLSVD) water suppression. methods were evaluated on simulated models and in vivo whole-brain MRSI using NMRSE, SNR, CRLB, and FWHM metrics. Results: WALINET is significantly faster and needs 8s for high-resolution whole-brain MRSI, compared with 42min for conventional HLSVD+L2. WALINET suppresses lipid and water in the brain by 25-45 and 34-53-fold, respectively. WALINET has better performance than HLSVD+L2, providing: (1) more lipid removal with 41% lower NRMSE;(2) better metabolite signal preservation with 71% lower NRMSE in simulated data;155% higher SNR and 50% lower CRLB in in vivo data. Metabolic maps obtained by WALINET in healthy subjects and patients show better gray-/white-matter contrast with more visible structural details. Conclusions: WALINET has superior performance for nuisance signal removal and metabolite quantification on whole-brain 1H-MRSI compared with conventional state-of-the-art techniques. This represents a new application of deep learning for MRSI processing, with potential for automated high-throughput workflow.
The widely employed Convolutional neural Networks (CNNs) encounter a trade-off between model complexity and performance in low-light image enhancement tasks. The current stage of Transformer performs well in low-light...
详细信息
The widely employed Convolutional neural Networks (CNNs) encounter a trade-off between model complexity and performance in low-light image enhancement tasks. The current stage of Transformer performs well in low-light image enhancement tasks, but its complexity restricts its application in real-time imageprocessing. To address this issue, this paper proposes a novel real-time and efficient low-light image enhancement network called RTEE-Net (Real-time Efficient Enhancement Network). RTEE-Net employs a Shallow Feature Enhancement Module (SFEM) that combines dense and residual connections to effectively learn more expressive shallow features. Subsequently, Deep Detail Enhancement Block (DDEB) is able to keep the image resolution unchanged while processing the image, preserving more high-frequency detailed texture information in the image. Furthermore, brightness adjustment suggestions generated through high-order curve adjustment methods provide global information to the network. While, Holographic Attention (HA) is introduced to better fuse global brightness and detail information, addressing issues such as color deviation and detail loss arising from improper information fusion. The experimental results show that RTEE-Net not only performs well in terms of image noise reduction, color artifact suppression, and color fidelity, but also demonstrates competitive performance in terms of real-time performance. Specifically, RTEE-Net contains only 50k parameters and has an average processing time of 0.002 s per image, and achieves a PSNR of 26.393 and a SSIM score of 0.863 in the LOL-v1 dataset.
This paper introduces a novel class of multivariate neural network operators activated by smooth ramp functions. Building upon the work of Qian and Yu (2022a), we explore the approximation properties of these operator...
详细信息
This paper introduces a novel class of multivariate neural network operators activated by smooth ramp functions. Building upon the work of Qian and Yu (2022a), we explore the approximation properties of these operators in both continuous and discrete settings. Key Contributions: center dot Theoretical Analysis: We establish direct and inverse theorems for the approximation of functions in the space C(R) of continuous functions on a compact domain R subset of Rs,s is an element of N and the space & Laplacetrf;(p)(R) of p-Lebesgue integrable functions on R, where 1 <= pSobolev Space Approximation: We extend our analysis to Sobolev spaces, a significant contribution to the field of neural network approximation. center dot Numerical Validation: We conduct numerical experiments to validate the theoretical results and compare the performance of our operators with existing methods. We use metrics such as Mean Squared Error (MSE) and Peak signal-to-Noise Ratio (PSNR) to evaluate the quality of the approximations. center dot Real-world Application: We demonstrate the practical applicability of our operators in imageprocessing, specifically in image compression and decompression. Flowcharts are provided to visualize the numerical experiments and the image compression/decompression process (Figure 5 and Figure 11). Our results show that the proposed operators offer improved approximation capabilities compared to existing neural network operators, as evidenced by the lower MSE and higher PSNR values obtained in our experiments.
This article presents a methodology for fault localization in electric power distribution systems through the analysis of wave matrix image using Convolutional neural Networks (CNN). Ensuring a continuous and high-qua...
详细信息
This article presents a methodology for fault localization in electric power distribution systems through the analysis of wave matrix image using Convolutional neural Networks (CNN). Ensuring a continuous and high-quality supply of electric power is crucial for the efficient operation of a Power System. However, the extensive coverage of the Electric System makes it susceptible to various disturbances caused by factors such as adverse weather conditions, equipment failures, presence of animals in the networks, and human errors. These disturbances can result in faults, characterized as short-circuits or abnormal currents between conductors or to the ground. With approximately 80% of power supply interruptions attributed to Distribution System faults, the need for effective fault localization methods is evident. By extracting voltage signal characteristics at measurement points during disturbances and transforming them into a panoramic representation of the system in the form of an image, the panoramic image generation, derived from simulated short-circuit values at each system bus, provides a comprehensive visualization of the network, enabling precise fault localization. The results demonstrate the accuracy of the CNN method in fault localization. This approach offers scalability, efficient transformation into images, and precise fault localization in electric power distribution systems.
Convolutional neural networks (CNNs) have gained significant popularity in image denoising. In recent years, many CNN-based image denoising methods have been developed. However, some of these methods extract the noise...
详细信息
Convolutional neural networks (CNNs) have gained significant popularity in image denoising. In recent years, many CNN-based image denoising methods have been developed. However, some of these methods extract the noise from noisy images by only stacking numerous relatively common convolutional layers, which can easily lead to some issues, including more loss of image details and insufficient processing of complex real-world noisy images. To address these challenges more effectively, a new progressive multi-scale denoising network (PMSDNet) is proposed. Its denoising effectiveness is attributed to two key modules, namely progressive multi-scale fusion block (PMSFB) and pixel attention block (PAB), in which PMSFB enhances the noise extraction capability of PMSDNet by enlarging its receptive field and reducing important image feature information loss, and PAB further facilitates PMSDNet to effectively capture the noise and preserve more image details by guiding it to focus more on multi-noised pixels and high-frequency image regions. Experimental results demonstrate the inspiring denoising performance of PMSDNet in facing three types of image noises, i.e., non-blind synthetic noise, blind synthetic noise and real-world noise. Additionally, the PMSDNet is extended to single-image deraining, showing the promising deraining performance.
暂无评论