Search

In this paper, we propose a cascade AdaBoost neural network (CANN) based on concepts and construct of AdaBoost neurons and cascade structure. Compared with AdaBoost, CANN can represent complex relationships between features. In CANN, representation learning is performed through AdaBoost, and the method of random selection features is utilized to encourage the diversity of AdaBoost neurons. Through the cascade structure, CANN has the context structure for complex feature representation. At the same time, in order to avoid the problem of feature disappearance, shortcut connection is used to add the previous information to the later nodes. Furthermore, particle swarm optimization (PSO) algorithm is utilized to optimize the structure of CANN, it can obtain the number of iterations to achieve better performance. Two types of CANN are proposed based — binary-classification CANN (BCANN) or multi-classification CANN (MCANN). The performance of CANN is evaluated with two kinds of data sets: machine learning data sets and atrial fibrillation data set. A comparative analysis illustrates that the proposed CANN leads to better performance than the models reported in the literature. [ABSTRACT FROM AUTHOR]

As the core of a blockchain system, the consensus mechanism not only helps to maintain the consistency of node data, but also gets involved in the issuance of tokens and prevention of attacks. Since the first blockchain system was born, it has been continuously improved with the development of blockchain technology and evolved into multiple new branches. Starting with the basic introduction of consensus and the classic Byzantine Generals Problem in distributed computing area, this survey utilizes a thorough classification to explain current consensus protocols in the blockchain system, presents the characteristics of mainstream protocols (PoW, PoS, DPoS, PBFT, etc.) and analyzes the strengths and weaknesses of them. Then we evaluate the performance qualitatively and quantitatively. In the end, we highlight several research directions for developing more practical consensus protocols for the future. [ABSTRACT FROM AUTHOR]

The Internet of Vehicles (IoV) is employed to gather real-time traffic information for drivers, and base stations in 5G systems are used to assist in traffic data transmission. For rapid implementation, the applications in vehicles are available to be offloaded to edge nodes (ENs) which are enhanced from micro base stations. Despite the benefits of IoV and ENs, the explosive growth of offloaded vehicle applications exceeds the capacity of ENs, causing the overload of fractional ENs. Therefore, it is necessary to offload the computing applications in overloaded ENs to other idle ENs, while it is a challenge to select appropriate offloading destination ENs. In this paper, we first consider edge computing framework for computation offloading in IoV under the architecture of 5G networks. We then formulate a multi-objective optimization problem to select suitable destination ENs, which aims to minimize the vehicle application offloading delay and offloading cost as well as realizing the load balance of ENs. Moreover, a computation offloading method for IoV, named COV, is designed to solve the multi-objective optimization problem. Finally, various simulation analyses demonstrate the effectiveness and efficiency of COV. [ABSTRACT FROM AUTHOR]

With the widespread application of mobile edge computing (MEC), MEC is serving as a bridge to narrow the gaps between medical staff and patients. Relatedly, MEC is also moving toward supervising individual health in an automatic and intelligent manner. One of the main MEC technologies in healthcare monitoring systems is human activity recognition (HAR). Built-in multifunctional sensors make smartphones a ubiquitous platform for acquiring and analyzing data, thus making it possible for smartphones to perform HAR. The task of recognizing human activity using a smartphone's built-in accelerometer has been well resolved, but in practice, with the multimodal and high-dimensional sensor data, these traditional methods fail to identify complicated and real-time human activities. This paper designs a smartphone inertial accelerometer-based architecture for HAR. When the participants perform typical daily activities, the smartphone collects the sensory data sequence, extracts the high-efficiency features from the original data, and then obtains the user's physical behavior data through multiple three-axis accelerometers. The data are preprocessed by denoising, normalization and segmentation to extract valuable feature vectors. In addition, a real-time human activity classification method based on a convolutional neural network (CNN) is proposed, which uses a CNN for local feature extraction. Finally, CNN, LSTM, BLSTM, MLP and SVM models are utilized on the UCI and Pamap2 datasets. We explore how to train deep learning methods and demonstrate how the proposed method outperforms the others on two large public datasets: UCI and Pamap2. [ABSTRACT FROM AUTHOR]

In order to increase the automatic quality control level in the textile industry, depending on the big data collected by the Internet of things of the textile factories, this paper proposes a novel visual saliency–based defect detection algorithm, which has the capability of automatically detecting defect in both nonpatterned and patterned fabrics. The algorithm employs the histogram features extracted from the saliency maps to detect the fabric defects. The algorithm involves three main steps: (1) saliency map generation to highlight the defective regions and suppress the defect‐free regions, (2) saliency histogram features extraction and selection to obtain the feature vectors that can effectively discriminate between the defective and defect‐free fabric images, and (3) fabric defect detection using a two‐class support vector machine classifier that has been trained using sets of feature vectors extracted from defective and defect‐free fabric samples. Experimental results show that our method yields accurate detections, outperforming other state‐of‐the‐art algorithms. [ABSTRACT FROM AUTHOR]

Abstract The identification of abnormal power consumption state is an important but difficult issue in power consumption. The State Grid Corporation of China's electric energy data acquisition system is only capable of acquiring power consumption big data collected by smart energy meter terminals. In view of this fact, this study presents a method for identifying abnormal power consumption state. First, the spectral distribution of eigenvalues of the covariance matrix of the high-dimensional random matrix of massive volumes of power consumption data is analyzed based on high-dimensional random matrix theory. Then, a power consumption big data-based abnormal power consumption state identification method is proposed based on the statistical properties of random matrices. Finally, simulations are performed based on actual power consumption data from Guizhou Province, China. The simulation results show that the proposed method can not only satisfy urgent requirements of power grids for visualization, timeliness, reliability and security but also provide a new approach for data-driven smart visual monitoring of power consumption. [ABSTRACT FROM AUTHOR]

Abstract Diabetic retinopathy (DR) is a common complication of diabetes and one of the major causes of blindness in the active population. Many of the complications of DR can be prevented by blood glucose control and timely treatment. Since the varieties and the complexities of DR, it is really difficult for DR detection in the time-consuming manual diagnosis. This paper is to attempt towards finding an automatic way to classify a given set of fundus images. We bring convolutional neural networks (CNNs) power to DR detection, which includes 3 major difficult challenges: classification, segmentation and detection. Coupled with transfer learning and hyper-parameter tuning, we adopt AlexNet, VggNet, GoogleNet, ResNet, and analyze how well these models do with the DR image classification. We employ publicly available Kaggle platform for training these models. The best classification accuracy is 95.68% and the results have demonstrated the better accuracy of CNNs and transfer learning on DR image classification. [ABSTRACT FROM AUTHOR]

The implementation of smart networks has made great progress due to the development of the Internet of Things (IoT). LoRa is one of the most prominent technologies in the Internet of Things industry, primarily due to its ability to achieve long-distance transmission while consuming less power. In this work, we modeled different environments and assessed the performances of networks by observing the effects of various factors and network parameters. The path loss model, the deployment area size, the transmission power, the spreading factor, the number of nodes and gateways, and the antenna gain have a significant effect on the main performance metrics such as the energy consumption and the data extraction rate of a LoRa network. In order to examine these parameters, we performed simulations in OMNeT++ using the open source framework FLoRa. The scenarios which were investigated in this work include the simulation of rural and urban environments and a parking area model. The results indicate that the optimization of the key parameters could have a huge impact on the deployment of smart networks. [ABSTRACT FROM AUTHOR]

Abstract: Spontaneous facial expression recognition is significantly more challenging than recognizing posed ones. We focus on two issues that are still under-addressed in this area. First, due to the inherent subtlety, the geometric and appearance features of spontaneous expressions tend to overlap with each other, making it hard for classifiers to find effective separation boundaries. Second, the training set usually contains dubious class labels which can hurt the recognition performance if no countermeasure is taken. In this paper, we propose a spontaneous expression recognition method based on robust metric learning with the aim of alleviating these two problems. In particular, to increase the discrimination of different facial expressions, we learn a new metric space in which spatially close data points have a higher probability of being in the same class. In addition, instead of using the noisy labels directly for metric learning, we define sensitivity and specificity to characterize the annotation reliability of each annotator. Then the distance metric and annotators' reliability is jointly estimated by maximizing the likelihood of the observed class labels. With the introduction of latent variables representing the true class labels, the distance metric and annotators' reliability can be iteratively solved under the Expectation Maximization framework. Comparative experiments show that our method achieves better recognition accuracy on spontaneous expression recognition, and the learned metric can be reliably transferred to recognize posed expressions. [Copyright &y& Elsevier]

• By analysing the motion magnitude of the videos, the proposed spatiotemporal interest points algorithm could eliminate a lot of superuous vehicle video at the edge. • The video segmentation algorithm based on the multi-modal linear features combination divides the video sequence into segments of interests, and then extracts the video clips from these segments. • An optimized YOLOv3 vehicle detection algorithm is presented based on edge computing, which greatly improves the detection speed and further meets the low latency and high accuracy requirements. • The extensive numerical experimental verification results show our methods may achieve better performance, compared with the existing algorithms for different stages of the redundancy elimination, video segmentation, key frame selection and vehicle detection. In the Internet of Things enabled intelligent transportation systems, a huge amount of vehicle video data has been generated and real-time and accurate video analysis are very important and challenging work, especially in situations with complex street scenes. Therefore, we propose edge computing based video pre-processing to eliminate the redundant frames, so that we migrate the partial or all the video processing task to the edge, thereby diminishing the computing, storage and network bandwidth requirements of the cloud center, and enhancing the effectiveness of video analyzes. To eliminate the redundancy of the traffic video, the magnitude of motion detection based on spatio-temporal interest points (STIP) and the multi-modal linear features combination are presented which splits a video into super frame segments of interests. After that, we select the key frames from these interesting segments of the long videos with the design and detection of the prominent region. Finally, the extensive numerical experimental verification results show our methods are superior to the previous algorithms for different stages of the redundancy elimination, video segmentation, key frame selection and vehicle detection. [ABSTRACT FROM AUTHOR]

This paper presents a novel, fully automatic method to grayscale images colorization using a scene guided neural network. In our proposed method, given a training set of both grayscale images and their corresponding color images, we first extract features of each grayscale pixel. These features, together with the corresponding RGB values of that pixel are input to train a colorization neural network for each given scene. To improve the performance of colorization, in both speed and results, we further classify the input and training images into different scene classes. We adopt a linear image classification method to generate a scene guided codebook and use it to determine the scene class of the input image. The preliminary colorization result is then generated by the corresponding trained neural network of the scene class of the input image. Finally, an image guided filter is used to refine colorized images. Inspired by the recent success in deep learning techniques which provide stabilizing modeling of large-scale medical image data, the proposed paper formulating the enhancement and colorization problem, so that colorization techniques can be directly used to ensure medical images with high quality. The experimental results on a broad range of images demonstrate that our method has better colorization performance as compared to that of the state-of-the-art algorithms. [ABSTRACT FROM AUTHOR]

• A novel deep and context-aware CNN structure for accurate and fast text detection • Hierarchically channel wise attention scheme with channel wise and multilayer features • Adopts frequency-based deep compression method to build a lightweight text detector With the significant power of deep learning architectures, researchers have made much progress on effectiveness and efficiency of text detection in the past few years. However, due to the lack of consideration of unique characteristics of text components, directly applying deep learning models to perform text detection task is prone to result in low accuracy, especially producing false positive detection results. To ease this problem, we propose a lightweight and context-aware deep convolutional neural network (CNN) named as CE-Text, which appropriately encodes multi-level channel attention information to construct discriminative feature map for accurate and efficient text detection. To fit with low computation resource of embedded systems, we further transform CE-Text into a lighter version with a frequency based deep CNN compression method, which expands applicable scenarios of CE-Text into variant embedded systems. Experiments on several popular datasets show that CE-Text not only has achieved accurate text detection results in scene images, but also could run with fast performance in embedded systems. [ABSTRACT FROM AUTHOR]

It is demonstrated that the deployment cost profiling varies for the various AI architecture possibilities, and that this cost should be taken into account when deploying and choosing the AI/ML solution. I The article by Narengerile et al. i proposes a new beam training approach via deep reinforcement learning to take use of machine learning technologies that will be used to build sixth-generation (6G) communication networks. [Extracted from the article]

In the Industrial Internet of Things (IIoT) era, we have billions of Internet-connected devices (also referred to as "things"), which range from unmanned/autonomous vehicles (e.g., unmanned aerial vehicle - UAV) to domestic robots to traffic sensors. The round-based blockchain time-stamping scheme is also designed to be scalable, in the sense that the constant number of bytes embedded into a blockchain is independent from the number of input documents. In conclusion, the papers included in this special section have contributed to different aspects of the blockchain literature, and hopefully benefit the readers interested in the potential application of blockchain. [Extracted from the article]

• Limited information in time domain results in limited performance of feature classification. • The particularity of EEG signal makes it difficult to measure. • The strong correlation of EEG signals makes it difficult to build feature extraction network. With the promotion of brain-computer interface technology, it is possible to study brain control system through EEG signals in recent years. In order to solve the problem of EEG signal classification effectively, a motor imagery classification algorithm based on recurrence plot convolution neural network is proposed. Firstly, EEG signals are preprocessed to enhance the signal intensity in the exercise interval. Secondly, time-domain and frequency-domain features are extracted respectively to construct the feature mode of recurrence plot. Finally, a new neural network is established to realize the accurate recognition of left and right movements. This research can also be transferred to other research fields. [Display omitted] [ABSTRACT FROM AUTHOR]

Internet of connected vehicles (IoV) are expected to enable intelligent traffic management, intelligent dynamic information services, intelligent vehicle control, etc. However, vehicles' data privacy is argued to be a major barrier toward the application and development of IoV, thus causing a wide range of attentions. Local differential privacy (LDP) is the relaxed version of the privacy standard, differential privacy, and it can protect users' data privacy against the untrusted third party in the worst adversarial setting. Therefore, LDP is potential to protect vehicles' data privacy in the practical scenario, IoV, although vehicles exhibit unique features, e.g., high mobility, short connection times, etc. To this end, in this paper, we first give an overview of the existing LDP techniques and present the thorough comparisons of these work in terms of advantages, disadvantages, and computation cost, in order to get the readers well acquainted with LDP. Thereafter, we investigate the potential applications of LDP in securing IoV in detail. Last, we direct several future research directions of LDP in IoV, to bridge the gaps between LDP researches and the privacy preservation in IoV. The originality of this survey is that it is the first work to summarize and compare the existing LDP research work and that it also does an pioneering work toward the in-depth analysis of the potential applications of LDP in privacy preservation in IoV. [ABSTRACT FROM AUTHOR]

Internet of Things (IoT) applications introduce a set of stringent requirements (e.g., low latency, high bandwidth) to network and computing paradigm. 5G networks are faced with great challenges for supporting IoT services. The centralized cloud computing paradigm also becomes inefficient for those stringent requirements. Only extending spectrum resources cannot solve the problem effectively. Mobile edge computing offers an IT service environment at the Radio Access Network edge and presents great opportunities for the development of IoT applications. With the capability to reduce latency and offer an improved user experience, mobile edge computing becomes a key technology toward 5G. To achieve abundant sharing, complex IoT applications have been implemented as a set of lightweight micro-services that are distributed among containers over the mobile edge network. How to produce the optimal collocation of suitable micro-service for an application in mobile edge computing environment is an important issue that should be addressed. To address this issue, we propose a latency-aware micro-service mashup approach in this paper. Firstly, the problem is formulated into an integer nonlinear programming. Then, we prove the NP-hardness of the problem by reducing it into the delay constrained least cost problem. Finally, we propose an approximation latency-aware micro-service mashup approach to solve the problem. Experiment results show that the proposed approach achieves a substantial reduction in network resource consumption while still ensuring the latency constraint. [ABSTRACT FROM AUTHOR]

With the rapid development and wide application of computer, camera device, network and hardware technology, 3D object (or model) retrieval has attracted widespread attention and it has become a hot research topic in the computer vision domain. Deep learning features already available in 3D object retrieval have been proven to be better than the retrieval performance of hand-crafted features. However, most existing networks do not take into account the impact of multi-view image selection on network training, and the use of contrastive loss alone only forcing the same-class samples to be as close as possible. In this work, a novel solution named Multi-view Discrimination and Pairwise CNN (MDPCNN) for 3D object retrieval is proposed to tackle these issues. It can simultaneously input multiple batches and multiple views by adding the Slice layer and the Concat layer. Furthermore, a highly discriminative network is obtained by training samples that are not easy to be classified by clustering. Lastly, we deploy the contrastive-center loss and contrastive loss as the optimization objective that has better intra-class compactness and inter-class separability. Large-scale experiments show that the proposed MDPCNN can achieve a significant performance over the state-of-the-art algorithms in 3D object retrieval. • Slice and Concat layers are used to concurrently handle multiple batches and views • Clustering scheme is utilized to make the MDPCNN more distinguishable. • A new loss is proposed to improve the learned feature discrimination. [ABSTRACT FROM AUTHOR]

Magnetic resonance imaging (MRI) has become a widely used medical imaging method. Affected by imaging mechanism, magnetic field inhomogeneity and other factors, MRI images are often interfered by non-Gaussian noise such as Rician noise and non-central chi-square distribution noise. However, the existing MRI denoising methods cannot effectively remove different kinds of noise, and the image is prone to blur and detail loss, even artifacts. Thus, this paper proposes an adaptive denoising algorithm for MRI based on Nonlocal Structural Similarity and Low-Rank Sparse Representation (NSS-LRSR). Different from the existing methods, it is a new paradigm to adaptively filter non-Gaussian noise of MRI and it has a good effect on both spatially stable and spatially varying Rician or non-central chi-square distribution noise. The forward variance stable transformation is used to correct the deviation caused by non-Gaussian noise, and then the non-local information is regrouped. And considering the sparse of similar image blocks, we use improved weighted kernel norm minimization to represent the non-local image blocks based on estimation of noise's standard deviation; thereby the processed image block are aggregated and outputted. Experimental results show that compared with the currently popular algorithms, the proposed NSS-LRSR achieves better results in PSNR and SSIM quantitative indexes. [ABSTRACT FROM AUTHOR]

Deep learning has brought a significant progress in medical image analysis. However, their lack of interpretability might bring high risk for wrong diagnosis with limited clinical knowledge embedding. In other words, we believe it's crucial for humans to interpret how deep learning work for medical analysis, thus appropriately adding knowledge constraints to correct the bias of wrong results. With such purpose, we propose Representation Group-Disentangling Network (RGD-Net) to explain the process of feature extraction and decision making inside deep learning framework, where we completely disentangle feature space of input X-ray images into independent feature groups, and each group would contribute to diagnose of a specific disease. Specifically, we first state problem definition for interpretable prediction with auto-encoder structure. Then, group-disentangled representations are extracted from input X-ray images with the proposed Group-Disentangle Module, which constructs semantic latent space by enforcing semantic consistency of attributes. Afterwards, adversarial constricts on mapping from features to diseases are proposed to prevent model collapse during training. Finally, a novel design of local tuning medical application is proposed based on RGB-Net, which is capable to aid clinicians for reasonable diagnosis. By conducting quantity of experiments on public datasets, RGD-Net have been superior to comparative studies by leveraging potential factors contributing to different diseases. We believe our work could bring interpretability in digging inherent patterns of deep learning on medical image analysis. • RGD-Net extracts fine-grained, group-disentangled disease representations for better interpretation and prediction. • Adversarial constraint avoids shortcut problems and promotes global minimum convergence in RGD-Net. • RGD-Net significantly improves classification accuracy and disentangles information in experiments. [ABSTRACT FROM AUTHOR]

• CDText detect texts of arbitrary shapes by encoding context information. • Feature extractor refines feature map with dilated context encoding blocks. • Transformer aggregates text features of detection boxes for instance segmentation. Scene text detection task aims to precisely locate text regions in natural scenes. However, the existing methods still face challenges in detecting arbitrary-shaped text, due to their limited feature representation capability. To alleviate this problem, we propose a scene text detector, i.e., CDText, based on structure of context-aware deformable transformer. Specifically, CDText firstly adopts different convolution kernel designs for feature extraction, which designs receptive fields with different size for multi-scale feature perception and fusion. Meanwhile, multi-head self-attention mechanism is used to strengthen the reasoning ability of CDText in a global sense, thus enhancing feature maps with abundant context information by extracting implicit relationship between multi-scale text features. Moreover, CDText designs a segmentation head to segment text instances of arbitrary shapes from rectangular detection boxes. Experiments show that CDText is superior to comparative methods in detection accuracy, achieving F -scores of 92.7, 81.9, and 82.9 on ICDAR2013, Total Text, and CTW-1500 datasets, respectively. [ABSTRACT FROM AUTHOR]

• We drew the lattice-based cryptographic primitive called NTRU into OT implementation for the first time, which assured the high computational performance and post-quantum property of our scheme. Moreover, since the parameters used in NTRU are technically configured, our protocol is scalable to some extent. • According to the IND-CCA2 model, we argued that the fastest oblivious transfer protocol known today [25] is fragile under active adaptive attack. We also found that this protocol is not semantic secure under decisional Diffie-Hellman assumption over the bilinear cyclic group. Contrarily, the above security flaws are eliminated in our scheme as proved under universal composability frame. • Our scheme is more applicable to heterogeneous and distributed environments thanks to the reusability of encoded data set. In addition, both the computation and communication performances of our protocol are preferable than that of Chou's scheme. The experiment results also imply that our protocol is adaptive to resource-constrained equipment. Security is a key concern in any IoT deployment, particularly if we have to take into consideration future attacks facilitated by the use of quantum computers. Therefore, in this paper, we present a post-quantum lightweight 1-out- n oblivious transfer (OT) protocol, based on the NTRU cryptographic primitive. Compared to the OT scheme proposed by Chou and Orlandi in LATINCRYPT 2015, our protocol is more suitable for deployment in heterogeneous and distributed environment, due to the reusability of encoded data set. Findings from the performance evaluation indicate that the proposed protocol outperforms that of Chou and Orlandi [13] protocol, in terms of computation and communication costs. [ABSTRACT FROM AUTHOR]

Abstract In the paper, the superconvergence of continuous Galerkin finite element methods (CGFEMs) for linear delay differential equations of neutral type is presented. By the orthogonal analysis method and under the suitable condition, it is proven that the finite element solution is superconvergent at the nodal points and Lobatto points. Numerical experiments further confirm the effectiveness and the superconvergence of the CGFEMs. [ABSTRACT FROM AUTHOR]

With the explosion of social data comes a great challenge called information overloading. To overcome this challenge, recommender systems are expected to support users in quickly accessing the appropriate content. However, cold-start users are a formidable challenge in the design of recommender systems because the conventional recommendation services are based on a single data source, namely, a single field. Considering the advantages of social-based and cross-domain approaches involving further additional data, we propose a cross-domain recommender system, including three approaches, based on multi-source social big data. The proposed approach is expected to effectively alleviate the issues of cold-start users by transferring user preferences from a related auxiliary domain to a target domain. Moreover, the transferred preferences are able to improve the diversity of recommendations. Through adequate evaluations based on an actual dataset in the book and music domains, it is shown that the accuracies of the three proposed approaches are significantly improved compared with the conventional recommender approaches, such as collaborative filtering and matrix factorization. In particular, the proposed approaches are available to provide cold-start users with highly effective recommendations. [ABSTRACT FROM AUTHOR]

Deformable contours are widely applied in medical image segmentation, which are usually derived from appearance cues in medical images. However, the performance of deformed contour is suppressed in ultrasonic image segmentation by the weak, misleading boundaries and the complex shapes of lesion regions. In this paper, a novel deformable contour model is proposed for segmenting ultrasound image sequences, which aims to utilize the powerful ability of deep learning network in learning of image features to help the deformable contour model resist weaknessses of ultrasound images. The deep learning network is designed as a densely connected siamese architecture. It trains a contrastive loss that serves as a boundary searching metric of a deformable contour to segment ultrasound image sequences. In this network, the densely residual blocks and the attention focused blocks are designed to make the network efficiently propagate features and focus on the lesion region, and the feature memory module stores and generates the prior features to aid the evolution of a deformable contour. Moreover, for resisting the impact of misleading or weak boundary, the shape similarity of lesion regions is used to as a shape prior and integrated into the framework of deformable contour to constrain the change of contours. The experimental results for the clinical ultrasound image sequences demonstrate that compared to the state-of-the-art methods, the proposed method can provide more accurate results in HIFU ultrasound images. [ABSTRACT FROM AUTHOR]

Rib segmentation based on chest X-ray images is essential in the computer-aided diagnosis systems of lung cancer, which serves as an important step in the quantitative analysis of various types of lung diseases. However, the traditional methods are unable to segment ribs effectively due to the unclear edges and overlapping regions in X-ray images. A novel rib segmentation framework based on Unpaired Sample Augmentation and Multi-Scale Network is presented in this paper, aiming to improve the accuracy of ribs segmentation with limited labeled samples. First, the algorithm learns pneumonia-related texture changes via unpaired chest x-ray images and generates various augmented samples. Then, a multi-scale network attempts to learn hierarchical features using global supervision. Finally, the refined segmentation result of each organ is achieved by using a deep separation module and a comprehensive loss function. Specifically, the hierarchical features can greatly improve the robustness of multi-organ segmentation networks. The complex multi-organ segmentation task with limited labeled data is simplified with the designed deep separation module. We justify the proposed framework through extensive experiments. It achieves good performance with DSC, Precision, Recall, and Jaccard of 88.03, 88.25, 88.36, and 79.02%, respectively. The DSC value increases nearly by 3% compared to other popular methods. The experimental results show that our algorithm presents better segmentation performance for the overlapping region and fuzzy region of multiple organs, which holds research value and prospects for application. [ABSTRACT FROM AUTHOR]

The transformational and spatial proximities are important cues for identifying inliers from an appearance based match set because correct matches generally stay close in input images and share similar local transformations. However, most existing approaches only check one type of them or both types consecutively with manually set thresholds, and thus their matching accuracy and flexibility in handling large-scale images are limited. In this paper, we present an efficient clustering based approach to identify match inliers with both proximities simultaneously. It first projects the putative matches into a joint transformational-spatial space, where mismatches tend to scatter all around while correct matches gather together. A mode-seeking process based on joint kernel density estimation is then proposed to obtain significant clusters in the joint space, where each cluster contains matches mapping the same object across images with high accuracy. Moreover, kernel bandwidths for measuring match proximities are adaptively set during density estimation, which enhances its applicability for matching different images. Experiments on three standard datasets show that the proposed approach delivers superior performance on a variety of feature matching tasks, including multi-object matching, duplicate object matching and object retrieval. [ABSTRACT FROM AUTHOR]

We design and implement a Bezier-based multipath routing algorithm that allows a given source node to send samples of data to a given sink node in large scale sensor networks. Multipath routing can distribute the energy load onto the multiple routes and thus increase the lifetime and quality of the network. It is important to stress the fact that evenly regulating the routing task among the more nodes of the network can also protect a node from failure considering that a node with heavy duty is likely to deplete its power quickly. On the contrary, all the traffic would be shipped along the shortest path routing, corresponding to the heavily congested path case, which in turn leads to overload of the nodes along the optimal routes between the sink and source pair and finally shortens the lifetime of the network. Also, multipath routing can increase end-to-end throughput and provide load balancing. Our view is that minimizing energy consumption while meeting acceptable latency for applications can lead to significant power savings. Our simulation results demonstrate that Bezier-based multipath routing approach significantly outperforms previous SWEEP, Tributaries and Deltas, and the shortest path approaches in terms of lifetime and latency. [ABSTRACT FROM AUTHOR]

Deep learning has shown remarkable performance in quantity of vision tasks. However, its large network generally requires quantity of samples to support sufficient parameters learning during training process. Such high request greatly reduces efficiency when applying on a small dataset with few samples. To alleviate this problem, we propose a novel data enhancement method for few-shot learning via a cutout approach and feature enhancement. After enhancement, the generated network not only produces distinguish feature map without collecting more samples, but also achieves advantage of feature representation with high efficiency for computing. Specifically, cutout approach is simple yet highly effective for image regulation, which enhances input image matrix by adding a fixed mask to improve robustness and overall performance of network. Afterward, we perform feature enhancement by proposing a feature promotion module, which uses characteristics of dilated convolution and sequential processing to improve feature representation ability, thus improving efficiency of the whole network. We conduct comparative experiments on both miniImageNet and CUB datasets, where the proposed method is superior to comparative methods in both 1-shot and 5-shot cases. [ABSTRACT FROM AUTHOR]

Edge computing enables portable devices to provide smart applications, and the indoor positioning technique offers accurate location-based indoor navigation and personalized smart services. To achieve the high positioning accuracy, an indoor positioning algorithm based on particle filter requires a large number of sample particles to approximate the probability density function, which leads to the additional computational cost and high fusion delay. Focusing on real-time and accurate positioning, an edge computing-enabled green multi-source fusion indoor positioning algorithm called APFP is proposed based on adaptive particle filter in this paper. APFP considers both pedestrian dead reckoning (PDR) signals in mobile terminals and the received signal strength indication (RSSI) of Bluetooth, and effectively merges the error-free accumulation of trilateral positioning and the accurate short-range positioning of PDR, which enables mobile terminals adaptively perform particle filter to reduce the computing time and power consumption while ensuring positioning accuracy simultaneously. Detailed experimental results show that, compared with the traditional particle filter algorithm and the map-constrained algorithm, the proposed APFP reduces fusion computing cost by 59.89% and 54.37%, respectively. [ABSTRACT FROM AUTHOR]

Point cloud representation is a challenge to extracting sufficient semantic information while ensuring that the sparsely point cloud spatial structure is complete. Benefiting from the Transformer network, recent studies have promoted the development of point cloud representation by extracting refined attention features based on global context. However, there is still undesired semantic information loss in the feature extraction stage. Hence, this paper proposes a novel architecture for 3D point cloud representation, namely Relation-Shape Transformer Network (RS-TNet), to address above problem while maintaining the merits of relation-shape embedding mechanism so as to generate rich and robust local semantic features. Specifically, RS-TNet can achieve coarse-to-fine grained semantic information coverage by integrating the global multi-head self-attention and local Relation-Feature extraction module simultaneously. Moreover, theoretical analysis demonstrates that RS-TNet can explicitly introduce the spatial relation of points by learning underlying shapes. In this way, extracted features are of more shape awareness and robustness. As a result, the proposed RS-TNet achieves 90.9% class accuracy and 85.6% Intersection-over-Union on ModelNet40 and ShapeNet datasets, respectively. Further, ablation experiments verify the effectiveness of our RS-TNet in point cloud classification and part segmentation tasks. [ABSTRACT FROM AUTHOR]

• Propose a Group-Disentangled Representation Learning framework (GDRL). • Introduce an implicit group-swap structure. • Extract linking relationship between semantical concepts of pathology and visual features. • Demonstrate that GDRL can significantly improve classification accuracy. Deep learning methods have shown significant performance in medical image analysis tasks. However, they generally act like "black box" without explanations in both feature extraction and decision processes, leading to lack of clinical insights and high risk assessments. To aid deep learning in envisioning diseases with visual clues, we propose a novel Group-Disentangled Representation Learning framework (GDRL). The key contribution is that GDRL completely disentangles latent space into disease concepts with abundant and non-overlapping feature related explanations, thus enhancing interpretability in feature extraction and decision processes. Furthermore, we introduce an implicit group-swap structure by emphasizing the linking relationship between semantical concepts of disease and low-level visual features, other than explicit explanations on general objects and their attributes. We demonstrate our framework on predicting four categories of diseases from chest X-ray images. The AUROC of GDRL on ChestX-ray14 for thoracic pathologic prediction are 0.8630, 0.8980, 0.9269 and 0.8653 respectively, and we showcase the potential of our framework in enhancing interpretability of the factors contributing to different diseases. [ABSTRACT FROM AUTHOR]

Recently, low-orbit satellite networks have gained lots of attention from the society due to their wide coverage, low transmission latency, and storage and computing capacity. Providing seamless connectivity to users in different areas is envisioned as a promising solution, especially in remote areas and for marine communication. However, when jointly used with terrestrial networks composing satellite-terrestrial networks, the satellite moving speed is much faster than the ground terminal, which can cause inconsistent service from a single satellite, and therefore lead to frequent satellite handover. Moreover, due to the dynamic and time slot visibility of satellites, the topology of an intersatellite changes frequently, which results in loops during satellite handover, thereby reducing the utilization of links. To address these problems, we propose a digital twin-assisted storage strategy for satellite-terrestrial networks (INTERLINK), which leverages the digital twins (DTs) to map the satellite networks to virtual space for better communication. Specifically, we first propose a satellite storage-oriented handover scheme to minimize the handover frequency by considering the limited access time and capacity constraints of satellites. Then, a multiobjective optimization problem is formulated to obtain the optimal satellite by genetic algorithm. Finally, considering the timing visibility of the satellite links, a digital twin-assisted intersatellite routing scheme is introduced to improve the quality of data delivery between satellites. Simulation results demonstrate that the proposed INTERLINK can reduce both handover times and average propagation delay compared with its counterparts. Meanwhile, benefitting from integrated DT, both the quality of data delivery and the delay of intersatellite links are considerably improved. [ABSTRACT FROM AUTHOR]

Currently, 6G is being proposed as a new next-generation communication architecture. The inclusion of satellite communication networks makes it possible to genuinely space-air-ground integrated networks, especially the combination of 6G with machine learning algorithms, which has attracted increasing attention. Of particular note is the combination of collaborative learning with the Internet of Everything, which improves the intelligence and collaboration of devices with more robust privacy protection, higher computing power, and greater energy savings. However, the airspace environment in 6G is extraordinarily dynamic and delay-sensitive for services, and traditional federation learning is a synchronous aggregation model with slow convergence. To address this problem, the paper proposes a hierarchical satellite-ground collaborative architecture with three roles: remote cloud center, orbital edge computing server, and data node. We use the advantages of orbital edge computing and low-orbit satellite network communication to manage the parameter transfer problem during the collaboration process. Meanwhile, we propose an asynchronous adaptive collaborative aggregation algorithm (AFLS) for the satellite-ground architecture, enabling the federated members to update the model parameters in time during the dynamic joining aggregation process. Experiments show that our architecture and algorithm excel in training performance with 95 % accuracy and fewer communication rounds in image classification when the roles are in good collaboration. [ABSTRACT FROM AUTHOR]

In recent years, the use of edge computing to solve the problem of limited resources in the IoV has attracted more and more attention. Vehicles can upload tasks to the edge servers within their radio range for computing cooperation and offloading. However, due to shared resources among edge servers, the uneven distribution of vehicles may lead to many problems such as uneven tasks distribution, unbalanced load, and low computing efficiency on different edge servers. On the other hand, most of the existing works are proposed to address the load unbalance issue of edge servers with the help of remote cloud or vehicle cloud, but still leading to resource wasting problems especially for edge servers with a light load. To address this issue, a task transfer scheme has been proposed among different edge servers in this paper. First, we designed a partitioned and hierarchical software-defined network architecture for IoV. After that, a load balancing model is proposed under this network architecture, in which the global controller can manage and schedule the local tasks from each edge server. Next, we proved that the task allocation problem for our load balancing model under the proposed network architecture is an NP-hard problem. To solve this problem, a Deep Q-Network (DQN) model is proposed to minimize the mean square deviation of loads among different edge servers. Numerical results show that our proposed load balancing model can significantly improve the resource utilization of edge servers as well as reduce the computational latency of vehicular tasks. [ABSTRACT FROM AUTHOR]

Diagnosis of tumor and definition of tumor borders intraoperatively using fast histopathology is often not sufficiently informative primarily due to tissue architecture alteration during sample preparation step. Confocal laser microscopy (CLE) provides microscopic information of tissue in real-time on cellular and subcellular levels, where tissue characterization is possible. One major challenge is to categorize these images reliably during the surgery as quickly as possible. To address this, we propose an automated tissue differentiation algorithm based on the machine learning concept. During a training phase, a large number of image frames with known tissue types are analyzed and the most discriminant image-based signatures for various tissue types are identified. During the procedure, the algorithm uses the learnt image features to assign a proper tissue type to the acquired image frame. We have verified this method on the example of two types of brain tumors: glioblastoma and meningioma. The algorithm was trained using 117 image sequences containing over 27 thousand images captured from more than 20 patients. We achieved an average cross validation accuracy of better than 83%. We believe this algorithm could be a useful component to an intraoperative pathology system for guiding the resection procedure based on cellular level information. [ABSTRACT FROM AUTHOR]

In this paper, we focus on the problem of 3D dynamic (4D) facial expression recognition. While traditional methods rely on building deformation models on high-resolution 3D meshes, our approach works directly on low-resolution RGB-D sequences; this feature allows us to apply our algorithm to videos retrieved by widespread and standard low-resolution RGB-D sensors, such as Kinect. After preprocessing both RGB and depth image sequences, sparse features are learned from spatio-temporal local cuboids. Conditional Random Fields classifier is then employed for training and classification. The proposed system is fully-automatic and achieves superior results on three low-resolution datasets built from the 4D facial expression recognition dataset – BU-4DFE. Extensive evaluations of our approach and comparisons with state-of-the-art methods are presented. [ABSTRACT FROM AUTHOR]

In many brain network studies, brain functional connectivity data is extracted from neuroimaging data and then used for disease prediction. For now, brain disease data not only has a small sample but also has the problem of high dimensional and nonlinear. Therefore, deep clustering on brain functional connectivity data is very challenging. To solve these problems, we propose a Soft-orthogonal Constrained Dual-stream Encoder with Self-supervised clustering network (SSCDE), which consists of a pretext task and downstream task, which can fully mine the effective information in brain disease data. In the pretext task, we use two brain disease data under the same category to do cross-domain learning to obtain effective information from the same dataset. In the downstream task, to reduce redundancy and avoid negative coding, we propose a soft-orthogonal constrained dual-stream encoder to encode features separately. At the same time, we use the pseudo labels given by the pretext task as prior information for self-supervised learning. We conduct validation on different brain disease recognition tasks, and the result have proved that the proposed framework has achieved good performance compared with the unsupervised clustering analysis algorithms. To our knowledge, this is the first cross-domain assisted recognition study on brain functional connectivity data. The code is available at https://github.com/hulu88/SSCDE. • We propose a self-supervised clustering framework for brain disease. • We propose a soft-orthogonal constrained dual-stream encoder. • We propose a pretext task based on unsupervised domain adaption. [ABSTRACT FROM AUTHOR]

Floods result in substantial damage throughout the world every year. Accurate predictions of floods can significantly alleviate casualties and property losses. However, due to the complexity of hydrology process especially in a city with complicated pipe network, the accuracy of traditional flood forecasting models suffer from the performance degradation with the increasing of required prediction period. In the work, based on the collected historical data of Xixian City, Henan Province, China, using the Internet of Things system (IoT) in 2011-2018, a Bidirectional Gated Recurrent Unit (BiGRU) multi-step flood prediction model with attention mechanism is proposed. In our model, the attention mechanism is used to automatically adjust the matching degree between the input features and output. Besides, we use a bidirectional GRU model, which can process the input sequence from two directions of time series (chronologically and antichronologically), then merge their representations together. Compared with the prediction model using Long Short Term Memory (LSTM), our method can generate better prediction result, as can be seen from the arrival time error and peak error of floods during multi-step predictions. • A BiGRU multi-step flood prediction model based on attention mechanism was proposed. • Model tests were designed and compared with relevant models based on LSTM neural network. • The results show that the model has smaller deterministic coefficient and RMSE. • Within the allowable error range, the prediction accuracy of our model can reach 100%. [ABSTRACT FROM AUTHOR]

Accurate segmentation of multiple targets, such as ribs, clavicles, heart, and lung fields, from chest X-ray images is crucial for diagnosing various lung diseases. Currently, mainstream deep learning methods heavily rely on fully annotated large-scale datasets. However, annotating all objects in chest X-ray images is a labor-intensive and time-consuming task. The publicly available partially annotated chest X-ray datasets have varying annotation objects and standards, and they are seldom utilized comprehensively in existing studies. To address these challenges, we propose A Multi-objective Segmentation Method for chest X-rays based on Collaborative Learning from Multiple Partially Annotated Datasets (called: MSM-CLMPAD). Our approach first utilizes an encoder constructed with densely connected blocks to extract multi-scale features from multiple partially annotated datasets. Then, leveraging the overlapping relationships among segmentation targets, we design a dual decoder guided by the attention mechanism. The novel attention-guided decoder effectively disentangles the features corresponding to various targets. Importantly, we propose an alternating training strategy for different datasets, facilitating collaborative learning of the same network model across datasets with diverse annotation targets. The experimental results performing on four public datasets demonstrate that our method achieves superior Dice and Jaccard coefficients compared to other popular methods, particularly for overlapping targets and unclear regions. We also explore the mutual influence of different targets in chest X-rays, offering a solution for interaction among partial datasets and further alleviating the difficulties of data annotation for multi-organ segmentation tasks. • This is the multi-organ segmentation framework for chest X-rays. • It stands out by collaborative learning on multiple partially annotated datasets. • We construct a shared encoder using densely connected blocks. • It incorporates a synergistic attention skip connection module and an attention-guided multi-scale feature selection module. • An attention oriented dual decoder structure addresses the challenge of low segmentation accuracy. [ABSTRACT FROM AUTHOR]

In the multi-object tracking process, a long-term tracking algorithm for traffic scene based on deep learning is proposed to handle several challenging problems, such as the complex variation of background illumination, change of pixel due to partial occlusion, cumulative error and the short-term disappearance of the target. Firstly, we train a CNN to identify and determine the target bounding box in a traffic scene. Secondly, we use a particle filter (PF) as the tracker to implement the preliminary multi-object tracking. Finally, the multi-object tracking trajectory then is generated by the mutual supervision of the PF tracker and CNN detector. In order to evaluate the experimental results, we use the forward-backward (FB) error of our tracker at a certain moment. The experimental results show that the method can track single and multi-objects in long-term tracking in real-time. For the situation of target disappearance and reappearance, the proposed algorithm can also recover its long-term tracking. [ABSTRACT FROM AUTHOR]

• We propose an approach to compensate the visual defects caused by metamorphopsia • Our approach enables interactive measurement of distortion in user's visual field • We compensate the warped visual field through a real-time processing of video streams • We conducted an experiment on 17 patients affected by metamorphopsia • The results show the proposed system is able to reduce visual field distortion Advances in Augmented Reality technologies and, particularly, the availability of video see-through enabled head mounted displays (HMD), are allowing to devise new strategies to help individuals with visual impairments in daily life. In this work, an approach is proposed to compensate a serious visual impairment, known as metamorphopsia, a vision disorder characterized by deformed images. The goal is to provide patients with a digitally restored visual field, through real-time processing of video see-through streams captured from the HMD. To this regard, we present two contributions, respectively, an interactive discrete modeling of patient's eye-specific vision distortion and a compensation of the latter by means of corresponding real-time counter-distortion of incoming frames. Our approach, indeed, maps each of the video streams acquired by the stereoscopic video see-through cameras aboard the headset on a 2D polygonal mesh which is then counter-warped by moving its vertices based on the previously built distortion model and then displayed, restored, on the HMD's screen. First user evaluations report promising results along with usability issues related to HMD technology. [ABSTRACT FROM AUTHOR]

Inspired by instance segmentation algorithms, researchers have proposed quantity of segmentation-based methods for text detection, achieving remarkable results on scene text with arbitrary orientation and large aspect ratios. Following their success, we believe cascade architecture and extracting contextual information in multiple aspects are powerful to boost performance on the basis of segmentation-based methods, especially in decreasing false positive texts in complex natural scene. Based on such consideration, we propose a multiple-context-aware and cascade CNN structure, which appropriately encodes multiple categories of context information into a cascade R-CNN framework. Specifically, the proposed method consists of two stages, i.e., feature generation and cascade detection. During the first stage, we define ISTK (Isolated Selective Text Kernel) module to refine feature map, which sequentially encodes channel-wise and kernel-size attention information by designing multiple branches and different kernel sizes in isolate form. Afterwards, we build long-range spatial dependencies in feature map via non-local operations. Built on contextual feature map, Cascade Mask R-CNN structure progressively refines accurate boundaries of text instances with multi-stage framework. We conduct comparative experiments on ICDAR2015 and 2017-MLT datasets, where the proposed method outperform comparative methods in terms of effectiveness and efficiency measurements. [ABSTRACT FROM AUTHOR]

The goal of Visual Question Answering (VQA) is to answer questions about images. For the same picture, there are often completely different types of questions. Therefore, the main difficulty of VQA task lies in how to properly reason relationships among multiple visual objects according to types of input questions. To solve this difficulty, this paper proposes a deep neural network to perform multi-modal relation reasoning in multi-scales, which successfully constructs a regional attention scheme to focus on informative and question-related regions for better answering. Specifically, we firstly design regional attention scheme to select regions of interest based on informative evaluation computed by a question-guided soft attention module. Afterwards, features computed by regional attention scheme are fused in scaled combinations, thus generating more distinctive features with scalable information. Due to designs of regional attention and multi-scale property, the proposed method is capable to describe scaled relationships from multi-modal inputs to offer accurate question-guided answers. By conducting experiments on VQA v1 and VQA v2 datasets, we show that the proposed method has superior efficiencies than most of the existing methods. • Multiscale design to describe nature of VQA in involving multiple objects. • Regional Attention to select informative question-related regions. • Three proper designed stages for multimodal fusion among textual question and visual image. [ABSTRACT FROM AUTHOR]

With the drastic development of Internet of things, the number of connected mobile users (MUs) is increasing at an unprecedented speed. The increasing popularity of MUs has triggered more and more new mobile applications. However, these applications are sensitive to latency, which inevitably increases pressure on MUs. Fortunately, computation offloading of mobile edge computing is becoming a promising technology that can improve quality of service for MUs. However, it becomes much difficult when there are multiple edge servers (ESs) near to the MU, even to the multiple MUs. On the other hand, as the resources of ESs are heterogeneous and finite, and thus it is challenge to design effective offloading strategies for multiple MUs. To tackle the above challenges, we firstly establish a multi-objective optimization model concerning time consumption and energy consumption of MUs, and resource utilization of ESs. Moreover, we devise an end-edge-cloud collaborative computing offloading method based on improved Strength Pareto Evolutionary Algorithm 2 for addressing this mode. Finally, compared to benchmark methods, numerous experiments have proved that our proposed method is effective and efficient and can be widely used for the scenario of multiple MUs and multiple heterogeneous ESs. [ABSTRACT FROM AUTHOR]

Recently, image captioning has achieved great progress in computer vision and artificial intelligence. However, language models still failed to achieve the desired results in high-level visual tasks. Generating accurate image captions for a complex scene that contains multiple targets is a challenge. To solve these problems, we introduce the theory of attention in psychology to image caption generation. We propose two types of attention mechanisms: The stimulus-driven and the concept-driven. Our attention model relies on a combination of convolutional neural network (CNN) over images and long-short term memory (LSTM) network over sentences. Comparison of experimental results illustrates that our proposed method achieves good performance on the MSCOCO test server. [ABSTRACT FROM AUTHOR]

Considering the different weights of various heterogeneity node importance in power grid and communication network and the possibility of the power grid island as well as the independent operation of the local area network, a more accurate method is proposed to evaluate the fragility of a cyber-physical system. According to the quo of grid construction and the background of cyber-physical fusion, the hierarchical system and distributed system model are established, respectively. Simulation results of conducting random attack and deliberate attack on two systems indicate that the proposed method is correct and better than others and it can identify the balance of the network structure. The distributed system is more robust than the hierarchical system under different weighting factors, while the robustness of the hierarchical system is more sensitive to the weighting factor. [ABSTRACT FROM AUTHOR]