Search

Inspired by the recent success of machine learning (ML), the concept of semantic communication introduced by Weaver in 1949 has gained significant attention and has become a promising research direction. Unlike conventional communication systems, semantic communication emphasizes the precise retrieval of conveyed meaning from the source to the receiver, rather than focusing on the accurate transmission of symbols. Thus, semantic communication can achieve a significant gain in source data compression, alleviate communication bandwidth pressure, and support new intelligent services, which is envisioned as a crucial enabler of future sixth-generation (6G) networks. In this review, we critically summarize the advances made in semantic information and semantic communications, including theory, architecture, and potential applications. Moreover, we deeply explore the major challenges in developing semantic communications and present the development prospects, aiming to prompt further scientific and industrial advances in semantic communications.

It is time-consuming to acquire complete data by fully phase encoding in two orthogonal directions along with one frequency encoding direction. Under-sampling in the 3D k-space is promising in accelerating such 3D MRI process. Although 3D under-sampling can be conducted according to predefined probability density, the density-based method is not optimal. Because of the large amount of 3D data and computational cost, it is challenging to perform data-driven and learning-based 3D under-sampling and subsequent 3D reconstruction. To tackle this challenge, this paper proposes a deep neural network called EEUR-Net, realized by optimizing specific under-sampling patterns for the fully sampled 3D k-space data. Innovatively, our under-sampling algorithm employs an end-to-end deep learning approach to optimize phase encoding patterns and uses a 3D U-Net for image reconstruction of under-sampled data. Through end-to-end training, we obtain an optimized 3D under-sampling pattern, which significantly enhances the quality of the reconstructed image under the same acceleration factor. A series of experiments on a knee MRI dataset demonstrate that, in comparison to standard random uniform, radial, Poisson and equispaced Cartesian under-sampling schemes, our end-to-end learned under-sampling pattern considerably improves the reconstruction quality of under-sampled MRI images. [ABSTRACT FROM AUTHOR]

To improve the prediction accuracy of milling tool wear, a prediction method based on Attention-LSTM is proposed. In the training phase, first, the data are pre-processed by truncation, downsampling, and the Hampel filtering method, and then features are extracted by the time domain, frequency domain, and time-frequency domain analysis methods. Second, a deep neural network is designed to describe the complex nonlinear function between features and tool wear. Last, aiming at the insufficient prediction accuracy due to the LSTM lacking feature extraction and enhancement, the Attention mechanism is introduced to optimize the model. The results suggest that this prediction method provides an efficient strategy for milling tool wear prediction. [ABSTRACT FROM AUTHOR]

Wearing masks has been a recommended protective measure due to the risks of coronavirus disease 2019 (COVID‐19) even in its coming endemic phase. Therefore, deploying a "smart mask" to monitor human physiological signals is highly beneficial for personal and public health. This work presents a smart mask integrating an ultrathin nanocomposite sponge structure‐based soundwave sensor (≈400 µm), which allows the high sensitivity in a wide‐bandwidth dynamic pressure range, i.e., capable of detecting various respiratory sounds of breathing, speaking, and coughing. Thirty‐one subjects test the smart mask in recording their respiratory activities. Machine/deep learning methods, i.e., support vector machine and convolutional neural networks, are used to recognize these activities, which show average macro‐recalls of ≈95% in both individual and generalized models. With rich high‐frequency (≈4000 Hz) information recorded, the two‐/tri‐phase coughs can be mapped while speaking words can be identified, demonstrating that the smart mask can be applicable as a daily wearable Internet of Things (IoT) device for respiratory disease identification, voice interaction tool, etc. in the future. This work bridges the technological gap between ultra‐lightweight but high‐frequency response sensor material fabrication, signal transduction and processing, and machining/deep learning to demonstrate a wearable device for potential applications in continual health monitoring in daily life. [ABSTRACT FROM AUTHOR]

Deep Learning has been successfully applied to reconstruct Magnetic Resonance (MR) images from undersampled k -space data to achieve the acceleration of MRI. However, most existing works focused on 2D MRI scans and perform reconstruction in a slice-by-slice manner due to the tremendous computation and memory cost of 3D MRI reconstruction, resulting in a blank in the research of 3D MRI acceleration. Providing more accurate and diagnostic information, 3D MRI requires excessively long scan time. To accelerate 3D MRI, in this work, we design a lightweight and powerful cascaded 3D network DIR3D utilizing a previously proposed computation-friendly data processing strategy. Specifically, we propose an efficient block called Dual-Domain Inter-Scale Mutual Reinforcement Block (DIRB) to fuse multi-scale features locally and globally with neglectable computation and memory costs, which enhances the representative ability of the network. To allow more flexibility, we further redesign the commonly used Data Consistency (DC) layer by introducing a learnable adaptor which enables the network to perform point-wise adaptive merge of the reconstructed and sampled k -space data while ensuring data consistency. We conduct comprehensive experiments on the Stanford MRIData and evaluate our DIR3D from multiple perspectives. When achieving the same acceleration factors, our proposed DIR3D consistently outperforms other state-of-the-art 2D methods at multiple subsampling masks, especially for highly undersampled data, which provides strong evidence for the superiority of our DIR3D for 3D MRI acceleration. Additionally, at the inference stage, our DIR3D can achieve a much higher reconstruction efficiency. • In this paper, we aim to overcome the huge computational burden of 3D reconstruction and explore the potential of rapid 3D Magnetic Resonance Imaging by designing an efficient pure 3D cascaded network (DIR3D). • We combine multi-scale feature fusion and dual-domain learning in an efficient way to enhance the network. • We upgrade the Data Consistency (DC) layer in the field of fast MRI reconstruction to help increase the self-compatibility of the resulting k-space. • Utilizing the correlations between slices, 3D reconstruction methods have greater potential to achieve high acceleration of Magnetic Resonance Imaging. • Performing a point-wise adaptive fusion of the reconstructed and the originally collected k-space data at sampled positions can help improve the reconstruction quality. [ABSTRACT FROM AUTHOR]

In this paper, we study the outage probability minimizing problem in a two-hop cooperative relay network. To reduce outage probability, existing studies propose many schemes for relay selection and power allocation, which are usually based on the assumption of exact channel state information (CSI). However, it is difficult to obtain perfect instantaneous CSI in practical situations where channel states change rapidly, and thus traditional methods would not perform well. Considering these factors, we turn to the emerging reinforcement learning (RL) methods for solutions. RL methods do not need any prior knowledge of CSI, but use neural network for approximation and decision after interacting with communication environment. Nevertheless, conventional RL methods, including most deep reinforcement learning (DRL) methods, cannot perform well when the search space is too large. In addition, non-stationarity is a common problem when using hierarchical reinforcement learning (HRL), which is caused by the changing behavior in different hierarchies. Therefore, we first propose a DRL framework with an outage-based reward function, which is then used as a baseline. Then, we further design an HRL framework and training algorithm. By decomposing relay selection and power allocation into two hierarchical optimization objectives, and combining on- policy and off-policy methods in the HRL framework, our method successfully address the sparse reward and non-stationary problem. Simulation results reveal that compared with traditional DRL method, the proposed HRL training algorithm can converge faster and reduce the outage probability by 8% in two-hop relay network with the same outage threshold. [ABSTRACT FROM AUTHOR]

Deep cascaded networks have been extensively studied and applied to accelerate Magnetic Resonance Imaging (MRI) and have shown promising results. Most existing works employ a large cascading number for the sake of superior performances. However, due to the lack of proper guidance, the reconstruction performance can easily reach a plateau and even face degradation if simply increasing the cascading number. In this paper, we aim to boost the reconstruction performance from a novel perspective by proposing a parallel architecture called DIRECTION that fully exploits the guiding value of the reconstruction residual of each subnetwork. Specifically, we introduce a novel Reconstruction Residual-Based Feature Modulation Mechanism (RRFMM) which utilizes the reconstruction residual of the previous subnetwork to guide the next subnetwork at the feature level. To achieve this, a Residual Attention Modulation Block (RAMB) is proposed to generate attention maps using multi-scale residual features to modulate the image features of the corresponding scales. Equipped with this strategy, each subnetwork within the cascaded network possesses its unique optimization objective and emphasis rather than blindly updating its parameters. To further boost the performance, we introduce the Cross-Stage Feature Reuse Connection (CSFRC) and the Reconstruction Dense Connection (RDC), which can reduce information loss and enhance representative ability. We conduct sufficient experiments and evaluate our method on the fastMRI knee dataset using multiple subsampling masks. Comprehensive experimental results show that our method can markedly boost the performance of cascaded networks and significantly outperforms other compared state-of-the-art methods quantitatively and qualitatively. [ABSTRACT FROM AUTHOR]

Compared with 2D Magnetic Resonance Imaging (MRI), 3D MRI is more powerful for generating high resolution images and visualizing small anatomical structures. However, 3D MRI acquisition is much more time-consuming due to the significantly larger number of phase encoding steps, which is directly proportional to the acquisition time. This paper proposes to select a volume-adaptive small subset of phases to accelerate 3D MRI scans and accurately reconstruct 3D images from the corresponding undersampled 3D k-space data. To avoid the delays caused by computationally expensive yet high-performance volume-adaptive phase selection, we propose a strategy of selecting multiple phases based on sampled slices from the volume during idle time within the repetition time (TR). To enhance the performance of phase selection, we propose a novel three-directional cross-attention phase selection network. Additionally, to improve the reconstruction performance, we introduce a three-directional slice-wise volume reconstruction. To the best of our knowledge, the proposed method, which we called EVOLVE (l e arning vol ume-adapti ve phases), is the first work that learns volume-adaptive phases for fast 3D MRI. The extensive experimental results on a large-scale 3D MRI dataset at various acceleration factors demonstrate the substantial performance improvement in terms of image reconstruction achieved by using the EVOLVE method for phase selection compared to traditional learning free 3D MRI phase selection methods. [ABSTRACT FROM AUTHOR]

[Display omitted] • Cracks are monitored using distributed fiber optic sensor (DFOS) and deep learning. • A modified You Only Look Once (YOLO) model adequately interprets DFOS data. • Transfer learning is incorporated to improve the accuracy of the deep learning model. • The robustness of the proposed approach is evaluated in different test scenarios. • The mAP@0.5 of detecting spatially-distributed cracks reaches 0.968. Distributed fiber optic sensors (DFOSs) offer unique capabilities for crack monitoring via measuring strain distributions. However, manually interpreting strain distributions is labor-intensive and time-consuming. To address this challenge, this paper presents a deep learning approach for real-time automatic interpretation of strain distributions, aiming at monitoring spatially-distributed cracks. The proposed approach encompasses three key innovations. First, deep learning-based methods are developed to facilitate automatic detection and localization of spatially-distributed cracks. Second, transfer learning is incorporated to overcome the data scarcity issue in training deep learning models. This ensures robust performance even with limited data. Third, a split-and-merge method is developed, enhancing the accuracy of multi-crack detection. To evaluate the performance of the approach, experimental data from various cases were considered. The results demonstrate a mean average precision (mAP) of 0.968 for crack detection. The processing time for a set of DFOS data, containing 10,000 measurement points, was less than 0.05 s. [ABSTRACT FROM AUTHOR]

In massive multiple-input multiple-output (MIMO), it is much challenging to obtain accurate channel state information (CSI) after radio frequency (RF) chain reduction due to the high dimensions. With the fast development of machine learning(ML), it is widely acknowledged that ML is an effective method to deal with channel models which are typically unknown and hard to approximate. In this paper, we use the low complexity vector approximate messaging passing (VAMP) algorithm for channel estimation, combined with a deep learning framework for soft threshold shrinkage function training. Furthermore, in order to improve the estimation accuracy of the algorithm for massive MIMO channels, an optimized threshold function is proposed. This function is based on Gaussian mixture (GM) distribution modeling, and the expectation maximum Algorithm (EM Algorithm) is used to recover the channel information in beamspace. This contraction function and deep neural network are improved on the vector approximate messaging algorithm to form a high-precision channel estimation algorithm. Simulation results validate the effectiveness of the proposed network. [ABSTRACT FROM AUTHOR]

Recently, deep learning-based image deblurring and deraining have been well developed. However, most of these methods fail to distill the useful features. What is more, exploiting the detailed image features in a deep learning framework always requires a mass of parameters, which inevitably makes the network suffer from a high computational burden. We propose a lightweight fusion distillation network (LFDN) for image deblurring and deraining to solve the above problems. The proposed LFDN is designed as an encoder–decoder architecture. In the encoding stage, the image feature is reduced to various small-scale spaces for multi-scale information extraction and fusion without much information loss. Then, a feature distillation normalization block is designed at the beginning of the decoding stage, which enables the network to distill and screen valuable channel information of feature maps continuously. Besides, an information fusion strategy between distillation modules and feature channels is also carried out by the attention mechanism. By fusing different information in the proposed approach, our network can achieve state-of-the-art image deblurring and deraining results with a smaller number of parameters and outperform the existing methods in model complexity. [ABSTRACT FROM AUTHOR]

Acquiring incomplete k-space matrices is an effective way to accelerate Magnetic Resonance Imaging (MRI). It is an important and challenging task to accurately reconstruct images from such under-sampled k-space matrices. On the one hand, neither image-domain oriented nor frequency-domain oriented deep Convolutional Neural Networks can simultaneously employ both frequency features and spatial features for cooperatively improving reconstruction accuracy. On the other hand, existing dual-domain reconstruction methods adopt heavy encoder-decoder frameworks, resulting in low efficiency and information loss in the process of pooling. To deal with these problems, in this paper, we propose a full-resolution dual-domain reconstruction network, called DIIK-Net. The DIIK-Net consists of a full-resolution frequency-domain branch, a full-resolution image-domain branch, and cross-domain interaction modules between the two branches. The first novelty of the proposed method is that the features of each block of frequency-domain branch are extracted by 1 × 1 filters, which reduces computational cost and captures rich contextual information. Due to the fact that an element in frequency domain conveys information of the whole image, 1 × 1 convolutional blocks are able to extract large contextual information with the interaction of image domain. The second novelty is that the image-domain branch consists of a very small number of 3 × 3 convolutional blocks and each block has very large field of perception due to integration of frequency domain. The third novelty lies in the simple and effective cross-domain interaction module. Experimental results on the challenging fastMRI dataset demonstrate that the proposed method is capable of achieving higher reconstruction accuracy with a few number of parameters. [ABSTRACT FROM AUTHOR]

• A deep learning model is proposed for crop mapping from UAV hyperspectral data. • The model could make full use of spatial and spectral features simultaneously. • The model yields an accuracy of 86% and a Kappa index of 0.8347. • The dataset, UAV-HSI-Crop, has been released to promote future studies. UAV hyperspectral imagery (HSI) has the unique merits of both a very high spatial and spectral resolution, which provides a high-quality data source for automatic crop mapping. Recently, deep learning has been widely used in crop classification, however, the design of an accurate crop mapping model for HSI data still remains a challenging task. Therefore, this paper aims to propose a novel semantic segmentation model (HSI-TransUNet) for crop mapping, which could make full use of the abundant spatial and spectral information of UAV HSI data simultaneously. Specifically, the proposed HSI-TransUNet belongs to an improved version of TransUNet, and we have made four important modifications for HSI data. Firstly, a spectral-feature attention module is designed for spectral features aggregation in the encoder. Afterwards, a series of Transformer layers with residual connections are designed to learn global contextual features. In the decoder part, sub-pixel convolutions are adopted to avoid the chess-board effect in the segmentation results. Finally, we design a hybrid loss function to further refine the predictions for boundaries. Experiment results indicate that the proposed HSI-TransUNet has achieved good performance in crops identification with an overall accuracy of 86.05%. Ablation studies have been conducted to verify the effectiveness of each refined module in the HSI-TransUNet. Comparison experiments also show that HSI-TransUNet has outperformed several previous semantic segmentation models. The dataset in this paper, UAV-HSI-Crop, is publicly available. http://doi.org/10.57760/sciencedb.01898. [ABSTRACT FROM AUTHOR]