Video de-raining is an important yet highly challenging task in the field of computer vision. Though numerous video de-raining methods are developed with encouraging performance, two major challenges for video de-raining are still unsatisfactorily solved and need to be further investigated as follows: 1) how to sufficiently explore the useful spatio-temporal information from adjacent rainy frames to facilitate the rain removal, and 2) how to well preserve background details even in a video with illumination variance. Regarding the above challenges, this paper specifically develops a new two-stage video de-raining method, which cleverly integrates two typical modules that are beneficial for the video de-raining task, namely Spatio-Temporal Fusion (STF) module and Illumination-Invariant Detail Preservation (IIDP) module. The STF module is designed to fuse the spatio-temporal information from successive frames effectively, while the IIDP module is developed to deliver the enhanced features from the first stage sub-network to the second stage sub-network to preserve clear edge details of objects. Experimental results demonstrate the superiority of our proposed method over previous state-of-the-arts. The code will be publicly available at https://github.com/mapleTan1113/TSVDN.

The Industrial Internet of Things (IIoT) has been viewed as a typical application for the fifth generation (5G) mobile networks. This article investigates the energy efficiency (EE) optimization problem for the Device-to-Device (D2D) communications underlaying unmanned aerial vehicles (UAVs)-assisted IIoT networks with simultaneous wireless information and power transfer (SWIPT). We aim to maximize the EE of the system while satisfying the constraints of transmission rate and transmission power budget. However, the designed EE optimization problem is nonconvex involving joint optimization of the UAV’s location, beam pattern, power control, and time scheduling, which is difficult to tackle directly. To solve this problem, we present a joint UAV location and resource allocation algorithm to decouple the original problem into several subproblems and solve them sequentially. Specifically, we first apply the Dinkelbach method to transform the fraction problem to a subtractive-form one and propose a mulitiobjective evolutionary algorithm based on decomposition (MOEA/D)-based algorithm to optimize the beam pattern. We then optimize UAV’s location and power control using the successive convex optimization techniques. Finally, after solving the above variables, the original problem can be transformed into a single-variable problem with respect to the charging time, which is linear and can be tackled directly. Numerical results verify that significant EE gain can be obtained by our proposed algorithm as compared to the benchmark schemes.

Reconfigurable intelligent surface (RIS) is a promising technique that smartly reshapes wireless propagation environment in the future wireless networks. In this paper, we apply RIS to an unmanned aerial vehicle (UAV)-assisted non-orthogonal multiple access (NOMA) network, in which the transmit signals from multiple UAVs to ground users are strengthened through RIS. Our objective is to minimize the power consumption of the system while meeting the constraints of minimum data rate for users and minimum inter-UAV distance. The formulated optimization problem is non-convex by jointly optimizing the position of UAVs, RIS reflection coefficients, transmit power, active beamforming vectors and decoding order, and thus is quite hard to solve optimally. To tackle this problem, we divide the resultant optimization problem into four independent subproblems, and solve them in an iterative manner. In particular, we first consider the sub-solution of UAVs placement which can be obtained via the successive convex approximation (SCA) and maximum ratio transmission (MRT). By applying the Gaussian randomization procedure, we yield the closed-form expression for the RIS reflection coefficients. Subsequently, the transmit power is optimized using standard convex optimization methods. Finally, a dynamic-order decoding scheme is presented for optimizing the NOMA decoding order in order to guarantee fairness among users. Simulation results verify that our designed joint UAV deployment and resource allocation scheme can effectively reduce the total power consumption compared to the benchmark methods, thus verifying the advantages of combining RIS into the multi-UAV assisted NOMA networks.

Multi-scale architectures and attention modules have shown effectiveness in many deep learning-based image de-raining methods. However, manually designing and integrating these two components into a neural network requires a bulk of labor and extensive expertise. In this article, a high-performance multi-scale attentive neural architecture search (MANAS) framework is technically developed for image de-raining. The proposed method formulates a new multi-scale attention search space with multiple flexible modules that are favorite to the image de-raining task. Under the search space, multi-scale attentive cells are built, which are further used to construct a powerful image de-raining network. The internal multi-scale attentive architecture of the de-raining network is searched automatically through a gradient-based search algorithm, which avoids the daunting procedure of the manual design to some extent. Moreover, in order to obtain a robust image de-raining model, a practical and effective multi-to- one training strategy is also presented to allow the de-raining network to get sufficient background information from multiple rainy images with the same background scene, and meanwhile, multiple loss functions including external loss, internal loss, architecture regularization loss, and model complexity loss are jointly optimized to achieve robust de-raining performance and controllable model complexity. Extensive experimental results on both synthetic and realistic rainy images, as well as the down-stream vision applications (i.e., objection detection and segmentation) consistently demonstrate the superiority of our proposed method. The code is publicly available athttps://github.com/lcai-gz/MANAS.

Conventional face anti-spoofing methods might be poorly generalized to unseen data distributions. Thus, we improve the generalization of spoof detection from the multi-domain feature disentanglement. Specially, a two-branch convolutional network is proposed to separate spoof-specific features and domain-specific features from face images explicitly. The spoof-specific features are further used for live vs. spoof classification. To minimize correlation among these two features, we present a cross-adversarial training scheme, which requires each branch to act as adversarial supervision for the other branch. To further exploit the subdomains from source data, a mixing augmentation approach is proposed based on mixing domain-specific feature statistics from different instances. It ensures more abundant domain discrepancy and facilitates the disentanglement process. The proposed approach shows promising generalization capacity in several public face anti-spoofing datasets.

Rain removal is a highly demanding task because a rainy image in computer lacks discriminative information to distinguish the image details from the rain streaks. In this paper, we present a new High-Low-Frequency Guided De-raining (HLFGD) method to remove the rain streaks clearly while reserve the image details. Specifically, the proposed HLFGD is built with three network branches, namely global-structure branch, de-raining branch, and edge-detail branch, which achieve the collaboration by concatenating intermediate features. Among them, the global-structure and edge-detail branches aim to explore the high-low frequency information, and the de-raining branch leverages the resulting spatial frequency information to restore the global structure of image and to retain fine edge details of objects during the de-raining process. Besides, a new architecture unit, called Residual Co-ordinate Attention Block (RCAB), is proposed to improve the effect of rain removal. Experimental results show the superiority of our method for image de-raining quantificationally and qualitatively.

Most recognizing occluded faces methods focus on synthetic-occluded faces for training due to the lack of real-occluded data. However, the performance may suffer from degradation since the synthetic-occluded and real-occluded face images are under different distributions. Hence, it draws our eyes to transfer the model from the synthetic to the real-world domain. In this paper, we propose a source data-free domain adaptive occluded face recognition framework to optimize the network in the target domain via redefining it as a pseudo labels denoising problem. To obtain reliable pseudo labels, we train synthetic-occluded and non-occluded images via distribution alignment to extract occlusion-robust features. Nonetheless, completely correct labels are still unattainable. Then, a denoising strategy is proposed to optimize pseudo labels by centroid-based feature clustering. Experiments show that the proposed approach can effectively recognize the real-occluded face; it also reminds the occluded faces recognition community about the feasibility of domain adaptation in existing tasks.

Single image de-raining is an important and highly challenging problem. To address this problem, some depth or density guided single-image de-raining methods have been developed with encouraging performance. However, these methods individually use the depth or the density to guide the network to conduct image de-raining. In this paper, a noveljoint depth and density guided de-raining(JDDGD) method is technically developed. The JDDGD starts with adepth-density inference network(DDINet) to extract the depth and density information from an input rainy image, followed by adepth-density-basedconditional generative adversarial network (DD-CGAN) to exploit the depth and density information provided by DDINet to achieve adaptive rain streak and fog removal. To prevent the spatially-varying local artifacts, an effectiveglobal-local discriminatorsstructure is introduced in the proposed DD-CGAN to globally and locally inspect the generated images. In addition, multiple loss functions includingmulti-scale pixel loss,multi-scale perceptual loss, andglobal-local generative adversarial lossare also jointly used to train our model to achieve the best performance. Both quantitative and qualitative results show that the proposed JDDGD method achieves superior performance than previousnon-guided,density-guided, anddepth-guided de-rainingmethods.

This paper investigates the energy efficiency (EE) optimization problem for device-to-device (D2D) communications underlaying non-orthogonal multiple access (NOMA) unmanned aerial vehicles (UAVs)-assisted networks with simultaneous wireless information and power transfer (SWIPT). Our aim is to maximize the energy efficiency of the system while satisfying the constraints of transmission rate and transmission power budget. However, the considered EE optimization problem is non-convex involving joint optimization of the UAV's location, beam pattern, power control and time scheduling, which is difficult to solve directly. To tackle this problem, we develop an efficient resource allocation algorithm to decompose the original problem into several sub-problems and solve them sequentially. Specifically, we first apply the Dinkelbach method to transform the fraction problem to a subtractive-form one, and propose a mulitiobjective evolutionary algorithm based on decomposition (MOEA/D) based algorithm to optimize the beam pattern. We then optimize UAV's location and power control by applying the successive convex optimization techniques. Finally, after solving the above variables, the original problem is transformed into a single-variable problem with respect to the charging time, which is a linear problem and can be solved directly. Numerical results verify that the significant EE gain can be obtained by our proposed method as compared to the benchmark schemes.

Most of the face anti-spoofing methods improve the generalization capability by adversarial domain adaptation via training the source and target domain data jointly. However, considering the data privacy, it is impractical in application. Hence, we propose a source data-free domain adaptative face anti-spoofing framework to optimize the network in the target domain without using labeled source data via modeling it into a problem of learning with noisy labels. To obtain more reliable pseudo labels, we propose dynamic images with the background to capture the motion divergences between real and attack faces. Nonetheless, fluctuations of predictions caused by noisy labels are still strong. Therefore, a filtering strategy is proposed to reduce the impact of noisy labels by self-ensemble, which combines prototype and progressive pseudo labels predicted by the source pre-trained model and target model respectively. The proposed approach shows promising generalization capability in several public-domains face anti-spoofing databases.

Abstract Stereo matching, which is a key problem in computer vision, faces the challenge of radiometric distortions. Most of the existing stereo matching methods are based on simple matching cost algorithms and appear the problem of mismatch under radiometric distortions. It is necessary to improve the robustness and accuracy of matching cost algorithms. A novel encoding pattern is proposed for stereo matching. In the proposed encoding pattern, each of the matching windows in the grey image and gradient images is divided into several isoline‐like sets with different radii. Then, pixel pairs are defined in the isoline‐like sets. An encoding function is used to decide the relative order between the two pixels in each pixel pair. To apply the pattern for matching cost computation and enhance the matching accuracy, an adaptively weighted cost is designed that is related to the isoline‐like sets. Experiments are conducted on the Middlebury and KITTI data sets to show the validity of the proposed method under severe radiometric distortions. Also, the comparisons with some widely used methods are made in the experiments to illustrate the advantage of the proposed method.

This paper considers an unmanned aerial vehicle (UAV)-aided millimeter Wave (mmWave) multiple-input-multiple-output (MIMO) non-orthogonal multiple access (NOMA) system, where a UAV serves as a flying base station (BS) to provide wireless access services to a set of Internet of Things (IoT) devices in different clusters. We aim to maximize the downlink sum rate by jointly optimizing the three-dimensional (3D) placement of the UAV, beam pattern and transmit power. To address this problem, we first transform the non-convex problem into a total path loss minimization problem, and hence the optimal 3D placement of the UAV can be achieved via standard convex optimization techniques. Then, the multiobjective evolutionary algorithm based on decomposition (MOEA/D) based algorithm is presented for the shaped-beam pattern synthesis of an antenna array. Finally, by transforming the original problem into an optimal power allocation problem under the fixed 3D placement of the UAV and beam pattern, we derive the closed-form expression of transmit power based on Karush-Kuhn-Tucker (KKT) conditions. In addition, inspired by fraction programming (FP), we propose a FP-based suboptimal algorithm to achieve a near-optimal performance. Numerical results demonstrate that the proposed algorithm achieves a significant performance gain in terms of sum rate for all IoT devices, as compared with orthogonal frequency division multiple access (OFDMA) scheme.

3D Morphable Model (3DMM) is a statistical tool widely employed in reconstructing 3D face shape. Existing methods are aimed at predicting 3DMM shape parameters with a single encoder but suffer from unclear distinction of different attributes. To address this problem, Two-Pathway Encoder-Decoder Network (2PEDN) is proposed to regress the identity and expression components via global and local pathways. Specifically, each 2D face image is cropped into global face and local details as the inputs for the corresponding pathways. 2PEDN is trained to predict 3D face shape components with two sets of loss functions designed to supervise 3D face reconstruction error and face identification error. To reduce the conflict between abundant facial details and saving computer storage space, a magnitudes converter is devised. Experiments demonstrate that the proposed method outperforms several 3D face recontruction methods.

We propose a novel joint framework for 3D face reconstruction (3DFR) that integrates facial attribute estimation (FAE) as an auxiliary task. One of the essential problems of 3DFR is to extract semantic facial features (e.g., Big Nose, High Cheekbones, and Asian) from in-the-wild 2D images, which is inherently involved with FAE. These two tasks, though heterogeneous, are highly relevant to each other. To achieve this, we leverage a Convolutional Neural Network to extract shared facial representations for both shape decoder and attribute classifier. We further develop an in-batch hybrid-task training scheme that enables our model to learn from heterogeneous facial datasets jointly within a mini-batch. Thanks to the joint loss that provides supervision from both 3DFR and FAE domains, our model learns the correlations between 3D shapes and facial attributes, which benefit both feature extraction and shape inference. Quantitative evaluation and qualitative visualization results confirm the effectiveness and robustness of our joint framework.

Accurate detection of the key components of transmission lines is an important part of smart grid construction. However, the detection of key components of transmission lines faces the problems of severe occlusion, irregular shape, and large size differences, which present a great challenge for anchor-based object detectors. We propose the anchor-based and anchor-free (ABAF) model, an object detection algorithm for both general and special object datasets. The ABAF detector is jointly trained with the anchor-based branch and anchor-free branch. At the time of inference, the predicted results of both are fused to yield the final detections. The experimental results show that the anchor-based branch is good at detecting shape-regular objects and the anchor-free branch is better at detecting irregularly shaped objects than anchor-based branch, and our fusion model ABAF has strong robustness for different datasets with excellent and stable performance. ABAF with ResNet-50 achieves 89.69% mAP on the transmission line dataset, a 3.72% improvement over one-stage anchor-based detector RetinaNet and we achieve 79.03% mAP on the PASCAL VOC dataset.

This paper develops an algorithm to effectively explore the advantages of both sparse vector recovery methods and generative model-based recovery methods for solving compressed sensing recovery problem. The proposed algorithm mainly consists of two steps. In the first step, a network-based projected gradient descent (NPGD) is introduced to solve a non-convex optimization problem, obtaining a preliminary recovery of the original signal. Then with the obtained preliminary recovery, a l1norm regularized optimization problem is solved by optimizing for sparse deviation vectors. Experimental results on two bench-mark datasets for image compressed sensing clearly demonstrate that the proposed recovery algorithm can bring about high computation speed, while decreasing the reconstruction error continuously with increasing the number of measurements.

As a prior knowledge, non‐local self‐similarity (NSS) has been widely utilised in ill‐posed problems. Actually, similar textures appear not only in a single scale, but also in different scales. Unlike most existing patch‐based methods that only explore NSS in the same scale, a multi‐scale patches based image denoising algorithm is proposed in this study. The authors have designed a multi‐scale strategy to expand the search space of block‐matching, which will increase the probability of finding more similar patches. After that, the weighted nuclear norm minimisation (WNNM) algorithm is employed to reveal latent clean patches. With the join of the multi‐scale framework, the performance of WNNM can be improved. The proposed algorithm can be used to solve NSS‐based image restoration tasks. In this study, mainly image denoising is studied, and its effectiveness is derived through experiments on widely used test images.

Quadrature spatial modulation (QSM) is one of the most prevalent transmission techniques for future wireless mobile network due to its high spectral efficiency (SE) and low complexity. However, it is restrictive to downlink of the cellular network deployments. Therefore, this paper proposes QSM for the multiuser uplink data transmission that increases the SE of the network by deploying multiple antennas at the mobile user with only two radio frequency (RF) chains. Maximum likelihood (ML) decoder is used to detect the received signal and attains the optimal detection performance but it is impractical because of its high computational complexity. Consequently, this paper proposes compressed sensing (CS) based orthogonal matching pursuit (OMP) detection as it has a suboptimal performance with a low computational complexity which can be a practical solution in the high‐density uplink multiuser network. However, conventional OMP algorithm has a low estimation performance due to multiuser interference which corrupt channel matrix. Thus, this work design an equalizer that mitigates the multiuser interference and improve the detection performance by orthonormalizing the columns of channel matrix. Simulation analysis confirm that the proposed QSM technique based on CS detector outperforms the conventional schemes in terms of SE and bit error rate (BER).

In this paper, a received signal strength indicator (RSSI) based indoor localization system is implemented employing Wi-Fi infrastructure. In the light of the feature-scaling-based k-nearest neighbor (FS-kNN) algorithm, a new continuous-feature-scaling model is proposed, which uses continuous weights instead of the discrete weights used in the FS-kNN, and needs not divide the entire RSSI space into the intervals. This gridless scheme avoids the difficulty of the weight selection at the common boundary of the adjacent intervals that could meet in the grid-based method of FS-kNN, which needs to divide the RSSI space into the intervals, ahead. An outlier deleting procedure is further used to improve the accuracy of the localization system. Experimental results indicate that the proposed method can be with a small localization error and is superior to some previous methods. One of the experiments achieves 1.34 m of the indoor localization error in a 12 m × 8 m area. The other is with 1.72 m of the indoor localization error in a 30 m × 25 m area. The proposed method performances best among the counterparts in the experiments.

Block‐sparsity is an extension of the ordinary sparsity in the realm of the sparse signal representation. Exploiting the block structure of the sparsity pattern, recovery may be possible under more general conditions. In this study, a block version of the orthogonal matching pursuit with thresholding (block‐OMPT) algorithm is proposed. Compared with the block version of the orthogonal matching pursuit (block‐OMP), block‐OMPT works in a less greedy fashion in order to improve the efficiency of the support estimation in iterations. Using the block restrict isometry property (block‐RIP), some performance guarantees of block‐OMPT are discussed for the bounded noise case and Gaussian noise case. A relationship between block‐RIP and block‐coherence is obtained. Numerical experiments are provided to illustrate the validity of the authors’ main results.

A new algorithm is proposed for compressed sensingmagnetic resonance imaging (CS-MRI). The lp-norm (0 <; p ≤ 1) based adaptive regularization model is used for MRI. The algorithm is established by using a novel iterative shrinkage scheme. In the iteration, the quasi-Newton method is employed. In the shrinkage, the threshold is defined varyingly. Also, the parameter p is selected dynamically in the algorithm. Comparing with some certain state-of-the-art methods for the noisy case, the proposed algorithm provides a higher accuracy of the MR image reconstruction. The performance of the proposed algorithm is validated by the theoretical analysis as well as some experimental results.

In this letter, robust sparse signal recovery is considered in the presence of the symmetric α-stable distributed noise. An M-estimate type model is constructed by approximating the location score function of the noise. A reweighed iterative hard thresholding algorithm is proposed to recover the sparse signal. The basis functions for the approximation and the recovery performance of the proposed algorithm are discussed. Simulations are given to demonstrate the validity of our results.

Spectrum access strategy plays a critical role in multichannel cognitive radio networks (CRNs). However, the CRNs cannot obtain the maximal throughput, when the existing access strategies, including overlay, underlay, and hybrid access strategies, are applied to multichannel CRNs. In this paper, we present a generalized access strategy in a multichannel CRN smart home environment, in which a secondary user (SU) system selects part of channels for sequential spectrum sensing, and accesses these channels based on the sensing results. Moreover, it accesses the remaining channels directly. We then formulate a two-phase optimization framework, which takes the sensing channel selection, sensing time allocation, and the power allocation into consideration, to maximize the gross average throughput of the multichannel CRN. In the sensing phase, a generalized access strategy algorithm (GAS) is first proposed, where we prove that only part of channels needs to be selected for spectrum sensing to achieve the maximum throughput. An optimal stopping rule is proposed to determine the optimal number of selected sensing channels. In addition, a completed hybrid access strategy algorithm is further investigated where the SU system senses all channels. An approximation algorithm is also presented to achieve suboptimal results with low computational complexity. In the transmission phase, the transmission powers of all channels are optimized via convex algorithms. Numerical experiments show that, compared with the existing schemes, the proposed schemes are able to achieve considerable throughput improvement.

Practically, in the underdetermined model Y = AX, where X is a K-group sparse matrix (i.e., it has no more than K nonzero rows), both Y and A could be totally perturbed. In this paper, based on restricted isometry property, for the greedy block coordinate descent algorithm, a sufficient condition of exact recovery is presented under the total perturbations, to guarantee that the support of the sparse matrix X is recovered exactly. It is pointed out that there exists some case satisfying our condition, but not the mutual coherence condition. We also discuss the upper bound of our sufficient condition.

In cognitive radio networks, spectrum sensing and access scheme affects the system performance. In this paper, a new wideband mixed access scheme is proposed, in which the Secondary Users (SUs) sense the channels via wideband spectrum sensing, and access them with a mixed access strategy. In order to maximize the ergodic throughput of SUs, we find optimal sensing time and transmission power of each channel, while protecting the Primary Users (PUs) from interference. It is shown that the optimization problem can be formulated as a convex problem. Moreover, we present a QoS-aware low complexity scheme, in which the SUs select several specific channels to sense. An effective sensing channels selection criterion is proposed. Numerical results show that the proposed schemes can effectively improve the system performance.

Block-sparse reconstruction, which arises from the reconstruction of block-sparse signals in structured compressed sensing, is generally considered difficult to solve due to the mixed-norm structure. In this letter, we propose an algorithm for reconstructing block-sparse signals, that is an extension of fixed point continuation in block-wise case by incorporating block coordinate descent technique. We also apply our algorithm to multiple measurement vector reconstruction, that is a special case of block-sparse reconstruction and can be used in magnetic resonance imaging reconstruction. Numerical results show the validity of our algorithm for both synthetic and real-world data.

This letter discusses blind separability based on temporal predictability (Stone, 2001 ; Xie, He, & Fu, 2005 ). Our results show that the sources are separable using the temporal predictability method if and only if they have different temporal structures (i.e., autocorrelations). Consequently, the applicability and limitations of the temporal predictability method are clarified. In addition, instead of using generalized eigendecomposition, we suggest using joint approximate diagonalization algorithms to improve the robustness of the method. A new criterion is presented to evaluate the separation results. Numerical simulations are performed to demonstrate the validity of the theoretical results.

This paper discusses the source recovery step in two-stage blind separation algorithm of underdetermined mixtures. A statistically non-sparse decomposition principle of two mixtures (2d-SNSDP), which is an extension of the SSDP algorithm about two mixtures, is proposed. It overcomes the disadvantage of the SSDP algorithm and sparse representation based on l1-norm. Compared with traditional sparse methods, it is non-sparse method, that is, almost all the recovered sources in any instant t are non-zero. Finally, several stereo audio signals experiments demonstrate its performance and practical.

Stone's method is one of the novel approaches to the blind source separation (BSS) problem and is based on Stone's conjecture. However, this conjecture has not been proved. We present a simple simulation to demonstrate that Stone's conjecture is incorrect. We then modify Stone's conjecture and prove this modified conjecture as a theorem, which can be used a basis for BSS algorithms.

Through an analysis and comparison of the algorithm proposed by Hyvarinen-Oja (1996), we present an approach of blind separation based on the penalty functions with multipliers. The approach gives the method to select the penalty function and speeds up the convergence of the algorithm. It avoids the ill-posed problem that may be caused by the pure penalty functions. Also, we propose a new simple method of deflation. The simulations show the good effectiveness of our algorithm. The separation time of our algorithm is shorter than the one proposed by Hyvarinen-Oja (1996) by 30%.

In this paper, a new algorithm of iterative learning control with forgetting factor is proposed by using a new norm and a new analysis method. The new method applies the whole information of systems to transfer the iterative learning control problem into a stability problem of a discrete system with parameters. This algorithm improves the shortage appeared in some present learning algorithms with forgetting factors. The simulations show the effectiveness of our new algorithm.