ABSTRACT Deepfake detection, as a defence against AI‐generated faces, has attracted significant attention. Existing image‐level detectors aim to mine forged traces in latent codes after pre‐trained backbones. However, merely considering such semantic‐level clues is often insufficient when confronted with unseen manipulations and datasets, where more complicated forgeries are encountered. To this end, this paper proposes a Multi‐Level Collaborations strategy, termed MLC, to enhance generalisation through simultaneously extracting pixel‐level fine‐grained, region‐level facial layout, and semantic‐level deep clues at different stages of encoding. Specifically, in the shallow stage, deformable convolutions with small receptive fields but adaptability, attached with spatial attentions, are used for spatial fine‐grained falsifies. In the middle stage, multiple dilated convolutions with different dilations in a pyramidal manner, further dynamically capture local incoordination within deepfakes. Finally, latent codes cooperated with such fine‐grained features, facilitate comprehensive discriminability via the long‐sequence dependency modelling system xLSTM. Moreover, multi‐task learning is employed for more stable multi‐level training. Extensive experiments show that MLC achieves superior performance compared to existing methods in both cross‐dataset and cross‐manipulation tests. The codes are available at: https://github.com/yanwd628/MLC .

Accurate detection and localisation of drones in dynamic battlefield environments remain challenging due to agile manoeuvres, occlusion effects, and unpredictable flight patterns. Traditional methods, such as vision-based tracking and Radio Frequency (RF) signal analysis, suffer from limited adaptability to real-time changes and high false-positive rates under cluttered conditions. To address these limitations, this paper proposes an Internet of Things-assisted Sparse Vector Detection Scheme (IoT-SVDS) that integrates recursive linear learning (RLL) and dynamic reference-position modelling. The IoT-SVDS first constructs a multi-dimensional feature vector (spatial coordinates, velocity, angular velocity) for each UAV, dynamically adjusted via adaptive filtering to mitigate occlusion and motion deviations. By fusing IoT sensor data and leveraging RLL, the system continuously refines prediction models without retraining, enabling rapid adaptation to novel flight patterns. Experimental validation in a simulated combat scenario (including RF jamming, sensor fusion, and hostile UAV swarms) demonstrates that IoT-SVDS outperforms state-of-the-art methods, achieving a 7.23% improvement in precision, an 11.07% reduction in localisation error, and 11.47% faster detection time compared to deep learning-based visual classifiers and RF-driven systems. Furthermore, the average trajectory deviation is reduced by 7.85%, validating its robustness in real-time battlefield communication and threat forecasting. These results highlight the potential of IoT-SVDS for military surveillance, autonomous engagement, and secure IoBT (Internet of Battlefield Things) applications. Future work will focus on enhancing the resilience and scalability of adversarial evasion in dense environments.

Existing reference-based blind face restoration (BFR) methods tend to either focus on the detailed texture of facial components or only conduct code prediction in the latent space, often neglecting their complementary relationship. To deal with it, the integration of local and non-local lookups that interact to ensure both fine-grained details and global geometric consistency is an eminently practical yet challenging solution. Specifically, Facial Component Dictionaries and High-Quality Feature Codebooks, which are pre-constructed from a large corpus of high-quality face images, perform their own functions of low- and high-level features. Thus, we first introduce an xLSTM-based Degradation-Aware Module (DAM) to mitigate code prediction biases suffering from degradation, then develop several two-stage Global Semantic Attention Modules (GSAM) to refine the multi-scale local details by the non-local insights. Finally, our proposed method, termed Local and Non-local Lookups for BFR (LNLFace), has demonstrated comparable or even superior performance than state-of-the-art methods, on both synthetic and real-world datasets. The codes are available at: https://github.com/yanwd628/LNLFace.

Facial image restoration has gained a tremendous progress since the increasing boom of the deep learning methods. Owing to its nature of strong ill-posedness, different categories of a-priori constraints have been harnessed or embedded in the existing deep architectures. While, as it turns to blind face restoration with more complicated degradations, the challenge becomes greater. In this paper, a further insightful step is taken by exploring the potentials of the graph convolutional networks (GCN) in conjunction with the structured priors for the blind problem. Specifically, a lightweight yet physically more intuitive model termed FaceGCN is proposed. On the one hand, a dynamic generator of facial adjacency matrices is constructed assisted by two self-supervised losses, allowing a sparse, accurate, and adaptive construction of case-specific face graphs with facial feature components as nodes. On the other hand, to model well the joint local-nonlocal correlations among various facial feature components, a kind of novel strip-attention GCN modules is correspondingly developed by splitting facial feature maps into intra- and inter-strips in both horizontal and vertical orientations, respectively. Extensive experimental results show that FaceGCN has achieved comparable or even superior performance to state-of-the-art methods, yet at a considerably less computational cost.

Traditional exclusive cloud resource allocation for deep learning training (DLT) workloads is unsuitable for advanced GPU infrastructure, leading to resource under-utilization. Fortunately, DLT workload co-location provides a promising way to improve resource utilization. However, existing workload co-location methods fail to accurately quantify interference among DLT workloads, resulting in performance degradation. To address this problem, this paper proposes an interference-aware scheduling approach for DLT workloads based on co-location performance prediction, dubbed ‘ISACPP’. ISACPP first builds an edge-fusion gated graph attention network (E-GGAT) that incorporates DL model structures, underlying GPU types, and hyper-parameter settings to predict co-location performance. Since the co-location state changes as each workload is completed, ISACPP proposes a multi-stage co-location interference quantification model derived from the predicted co-location performance to identify the GPU device with the minimum overall interference. Experimental results demonstrate that ISACPP can accurately estimate the co-location performance of DLT workloads with a maximum prediction error of 8.72%, 1.9%, and 4.4% for execution time, GPU memory consumption, and GPU utilization, respectively. Meanwhile, ISACPP can significantly shorten workload makespan by up to 34.9% compared to state-of-the-art interference-aware scheduling methods.

With the evolution of telemedicine, clean images, especially facial images, are crucial in areas such as symptom evaluation and cosmetic medicine. However, dealing with indiscernible facial images, a condition known as blind degradation, poses a formidable challenge for existing blind face restoration (BFR) techniques due to their inherently ill-posed nature. Thus, a delicate interaction between preserving local details and maintaining global geometries is desired. Despite advances in convolutional neural networks (CNNs), transformers, and denoising diffusion, where no matter serried convolutions, elaborately-designed self-attentions, or stochastic noises, tend to isolate either local or non-local features. To this end, a new candidate model termed Multi-Scale in Multi-Scale Mamba (M2Mamba) is proposed, which builds on a pioneering structured state space model-based network and includes three new components: multi-scale learned fusion module (MSLFM), multi-scale attention fusion module (MSAFM), and multi-scale inspired Mamba (MS-Mamba). Firstly, the MSLFM is adopted by extracting image-level global guidance from inputs of different scales, preserving intuitive perception yet enriching semantic understanding. Secondly, the MSAFM dynamically integrates features from various encoder stages. Thirdly, the MS-Mamba employs separate branches for both small- and large-scale receptive fields, benefiting modeling of long-range dependencies. In the final, M2Mamba is demonstrated on both synthetic and realistic benchmarks, showing comparable or better performance than state-of-the-art methods. The codes are available at: https://github.com/yanwd628/M2Mamba.

Video compressive sensing (VCS) presents a promising encoder paradigm for efficient video signals acquisition at resource-limited applications. In order to recover complete and accurate signals at the decoder, powerful reconstruction algorithms are desired to exploit rich temporal redundancies within video sequences. In this paper, we develop a flow-guided non-local alignment network (FNLAN), which can build accurate temporal dependencies among adjacent frames to help video recovery. The frames are aligned by non-local operations so that the subsequent fusion module can aggregate useful information from misaligned informative features. To avoid the prohibitive computation burden, the non-local operation is conducted within a local window guided by optical flows, i.e. the offsets derived from optical flows are enforced on the search window so that more correlated features are involved in non-local operations. Experimental results demonstrate the superiority of FNLAN.

Infrared and visible data fusion (IVF) aims to generate a fused output that simultaneously highlights salient thermal radiation features and preserves texture information, which can not only grasp the necessary information for traffic movement, but also highlight the invisible objects that need to be dodged in intelligent transportation system (ITS). Therefore, IVF is capable of improving the environmental perception ability for various challenging traffic situations, e.g., foggy scenarios, rainy environments, and low-light illumination. However, current available IVF algorithms cannot offer a theoretical manner to integrate a priori knowledge and the network structure into a unified model. Moreover, they always fail to handle infrared and visible data pairs with different resolutions, which is a common occurrence in real ITS scenarios. To this end, this study develops a novel model-inspired unsupervised network termed IVF-Net. Specifically, an enhanced IVF model (IVFM), which pays more attention on detailed texture information and salient objects, is first established. According to proximal gradient theory, then we map this model into a deep network with learnable feature extraction parameters, aiming to draw on the strengths of the fusion model and deep learning to better describe the IVF task. Finally, a multiple task-driven loss function is designed to train the mapped network. Unlike previous work, our IVF-Net is motivated by IVFM, each layer in which has a semantic interpretability and a clear mission, thereby leading to a significantly enhanced fusion effect. Another advantage is that it is only composed of simple convolution-based structures, which ensures its lightweight and efficiency. Experiments demonstrate that IVF-Net can have a stronger ability to capture the key traffic information and highlight the salient feature of imperceptible objects, which makes it an excellent candidate to improve the reliability of subsequent applications in ITS.

RGB-T tracker possesses strong capability of fusing two different yet complementary target observations, thus providing a promising solution to fulfill all-weather tracking in intelligent transportation systems. Existing convolutional neural network (CNN)-based RGB-T tracking methods often consider the multisource-oriented deep feature fusion from global viewpoint, but fail to yield satisfactory performance when the target pair only contains partially useful information. To solve this problem, we propose a four-stream oriented Siamese network (FS-Siamese) for RGB-T tracking. The key innovation of our network structure lies in that we formulate multidomain multilayer feature map fusion as a multiple graph learning problem, based on which we develop a graph attention-based bilinear pooling module to explore the partial feature interaction between the RGB and the thermal targets. This can effectively avoid uninformed image blocks disturbing feature embedding fusion. To enhance the efficiency of the proposed Siamese network structure, we propose to adopt meta-learning to incorporate category information in the updating of bilinear pooling results, which can online enforce the exemplar and current target appearance obtaining similar sematic representation. Extensive experiments on grayscale-thermal object tracking (GTOT) and RGBT234 datasets demonstrate that the proposed method outperforms the state-of-the-art methods for the task of RGB-T tracking.

Short-term traffic flow forecasting is a challenging subject, and it is of great significance for travel route planning, traffic regulation and other directions. Traffic flow is affected by the topological structure of the urban road network and the dynamic changes of time series, and has both temporal and spatial characteristics. However, how to extract the correlation between spatiotemporal features is still a challenging task. In response to these problems, this paper proposes a new deep learning model (GAGRU), which models the traffic road network through a graph attention network (GAT), extracts the spatial dependencies in the traffic flow and uses a gated recurrent unit (GRU) to focus on the Characteristics of traffic over time. In addition, we fuse the traffic flow features of multiple sequences to consider the periodic characteristics of traffic flow. In this paper, the model is experimentally validated using real-world datasets, and the final experimental results show that the prediction accuracy of the model is superior to other baseline methods.

Abstract The past decade witnessed a remarkable increase in deep learning (DL) workloads which require GPU resources to accelerate the training process. However, the existing coarse‐grained scheduling mechanisms are agnostic to information other than the number of GPUs or GPU memory, which results in performance degradation of DL tasks. Moreover, the common assumption held by the existing balance‐aware DL task scheduling strategies, a DL task consumes resources once it starts, fails to reduce resource contention, and further limits execution efficiency. To address these problems, this article proposes a fine‐grained and balance‐aware scheduling model (FBSM) which considers the resource consumption characteristic of the DL task. Based on FBSM, we propose customized GPU sniffer (GPU‐S) and balance‐aware scheduler (BAS) modules to construct a scheduling system called KubFBS. The experimental results demonstrate KubFBS accelerates the execution of DL tasks while improving the load balancing capability of the cluster.

Abstract Video super‐resolution (VSR) aims at recovering high‐resolution frames from their low‐resolution counterparts. Over the past few years, deep neural networks have dominated the video super‐resolution task because of its strong non‐linear representational ability. To exploit temporal correlations, most deep neural networks have to face two challenges: (1) how to align consecutive frames containing motions, occlusions and blurring, and establish accurate temporal correspondences, (2) how to effectively fuse aligned frames and balance their contributions. In this work, a novel video super‐resolution network, named NLVSR, is proposed to solve above problems in an efficient and effective manner. For alignment, a temporal‐spatial non‐local operation is employed to align each frame to the reference frame. Compared with existing alignment approaches, the proposed temporal‐spatial non‐local operation is able to integrate the global information of each frame by a weighted sum, leading to a better performance in alignment. For fusion, an attention‐based progressive fusion framework was designed to integrate aligned frames gradually. To penalize the points with low‐quality in aligned features, an attention mechanism was employed for a robust reconstruction. Experimental results demonstrate the superiority of the proposed network in terms of quantitative and qualitative evaluation, and surpasses other state‐of‐the‐art methods by 0.33 dB at least.

Atmospheric scattering model (ASM) is one of the most widely used model to describe the imaging processing of hazy images. However, we found that ASM has an intrinsic limitation which leads to a dim effect in the recovered results. In this paper, by introducing a new parameter, i.e., light absorption coefficient, into ASM, an enhanced ASM (EASM) is attained, which can address the dim effect and better model outdoor hazy scenes. Relying on this EASM, a simple yet effective gray-world-assumption-based technique called IDE is then developed to enhance the visibility of hazy images. Experimental results show that IDE eliminates the dim effect and exhibits excellent dehazing performance. It is worth mentioning that IDE does not require any training process or extra information related to scene depth, which makes it very fast and robust. Moreover, the global stretch strategy used in IDE can effectively avoid some undesirable effects in recovery results, e.g., over-enhancement, over-saturation, and mist residue, etc. Comparison between the proposed IDE and other state-of-the-art techniques reveals the superiority of IDE in terms of both dehazing quality and efficiency over all the comparable techniques.

We investigate an energy cost minimization problem for a smart home in the absence of a building thermal dynamics model with the consideration of a comfortable temperature range. Due to the existence of model uncertainty, parameter uncertainty (e.g., renewable generation output, nonshiftable power demand, outdoor temperature, and electricity price), and temporally coupled operational constraints, it is very challenging to design an optimal energy management algorithm for scheduling heating, ventilation, and air conditioning systems and energy storage systems in the smart home. To address the challenge, we first formulate the above problem as a Markov decision process, and then propose an energy management algorithm based on deep deterministic policy gradients. It is worth mentioning that the proposed algorithm does not require the prior knowledge of uncertain parameters and building the thermal dynamics model. The simulation results based on real-world traces demonstrate the effectiveness and robustness of the proposed algorithm.

Distributed compressive video sensing (DCVS) has great potential for signal acquisition and processing in source-limited communication, e.g., wireless video sensors networks, because it shifts complicated motion estimation and motion compensation from the encoder to the decoder. Known as a state-of-the-art technique in DCVS, multihypothesis (MH) prediction is widely used because of its acceptable performance and low computational complexity. However, this technique is restricted by inaccurate regularizations, which can cause susceptibility to inaccurate hypotheses. In this paper, we present an iterative reweighted Tikhonov-regularized scheme for MH prediction reconstruction. Specifically, to enhance robustness, this scheme proposes a reweighted Tikhonov regularization that synthetically considers three factors that affect the MH prediction performance—accuracy of the hypothesis set, number of hypotheses, and accuracy of regularizations—by utilizing the influence of each hypothesis. Furthermore, to avoid over-iteration in iterative MH prediction reconstruction, we propose a Bhattacharyya coefficient-based stopping criterion for use in the recovery of non-key frames, in which we exploit the similarity to an adjacent key frame rather than a previous iteration result. The simulation results show that the proposed scheme outperforms the state-of-the-art MH methods in terms of robustness to inaccurate hypotheses when there are a limited number of hypotheses.

This paper introduces a novel and effective image prior, i.e., gamma correction prior (GCP), which leads to an efficient image dehazing method, i.e., IDGCP. A step-by-step procedure of the proposed IDGCP is as follows. First, an input hazy image is preprocessed by the proposed GCP, resulting in a homogeneous virtual transformation of the hazy image. Then, from the original input hazy image and its virtual transformation, the depth ratio is extracted based on atmospheric scattering theory. Finally, a "global-wise" strategy and a vision indicator are employed to recover the scene albedo, thus restoring the hazy image. Unlike other image dehazing methods, IDGCP is based on the "global-wise" strategy, and it only needs to determine one unknown constant without any refining process to attain a high-quality restoration, thereby leading to significantly reduced processing time and computation cost. Each step of IDGCP is tested experimentally to validate its robustness. Moreover, a series of experiments are conducted on a number of challenging images with IDGCP and other state-of-the-art technologies, demonstrating the superiority of IDGCP over the others in terms of restoration quality and implementation efficiency.

This paper investigates the physical-layer security (PLS) of wireless transmissions with the aid of multiple decode-and-forward one-way relays in the presence of both eavesdropping attacks and channel estimation errors (CEEs). To protect wireless transmission with CEE, two opportunistic relaying schemes are conceived to upgrade the PLS of the wireless communications with the aid of relays, namely, the CEE-oriented pure relay selection (CEE-PRS) and CEE-oriented jammer-aided relay selection (CEE-JRS), respectively. Moreover, the security-reliability tradeoff (SRT) is designed due to that increasing the transmit power may enhance the reliability, but the security may be sacrificed concurrently, where the security and reliability are characterized by the intercept probability (IP) and outage probability (OP), respectively. We then analyze the IP and OP of the CEE-PRS and CEE-JRS schemes. Furthermore, both the CEE-oriented direct transmission (CEE-DT) and the perfect channel estimation-oriented direct transmission (PCE-DT) schemes are also analyzed for comparison purposes. It is shown that although the SRT of the wireless communications can indeed be degraded by the eavesdropping attacks in low CEE regions, the proposed CEE-PRS and CEE-JRS schemes are capable of significant improving the SRT performance of wireless communications, demonstrating the advantage of the proposed CEE-PRS and CEE-JRS schemes against eavesdropping.

Atmospheric scattering model (ASM) has been widely used in hazy image restoration. However, the recovered albedo might deviate from the real scene once the input hazy image cannot fully satisfy the model's assumptions such as the homogeneous atmosphere and even illumination. In this paper, we break these limitations and redefine a more reliable ASM (RASM) that is extremely adaptable for various practical scenarios. Benefiting from RASM, a simple yet effective Bayesian dehazing algorithm (BDPK) is further proposed based on the prior knowledge. Our strategy is to convert the single image dehazing problem into a maximum a-posteriori probability one that can be approximated as an optimization function using the existing priori constraints. To efficiently solve this optimization function, the alternating minimizing technique is introduced, which enables us to directly restore the scene albedo. Experiments on a number of challenging images reveal the power of BDPK on removing haze and verify its superiority over several state-of-the-art techniques in terms of quality and efficiency.

Distributed compressive video sensing (DCVS) is a novel video coding technique that shifts sophisticated motion estimation and compensation from the encoder to the decoder and is suitable for resource‐limited communication, namely wireless video sensor networks (WVSNs). In DCVS, key frames serve as the reference for subsequent non‐key frames in a given group of pictures (GOP). However, in fast‐motion sequences, e.g. scene‐changing sequences, a fixed GOP size can cause inaccuracy in the selection of reference key frames. The difference in the peak signal‐to‐noise ratio between key frames and non‐key frames caused by this inaccuracy appears as flicker in the decoded video, negatively affecting the quality of experience. To address this problem, the authors present a perceptual hash algorithm‐based adaptive GOP selection algorithm for DCVS and a novel allocation model for the frame sampling rate. In addition, the authors define several indexes to assess the degree of flicker in decoded video. The experimental results demonstrate that the proposed algorithm reduces the degree of flicker in fast‐motion sequences by 40–60% relative to the state‐of‐the‐art architecture, while also outperforming other adaptive GOP selection strategies.

In this letter, a concise gamma-correction-based dehazing model (GDM) is proposed. This GDM explicitly describes the inner relationship between the gamma correction (GC) and the traditional scattering model. Combined with the existing priori constraints, GDM is further approximated into a one-dimensional (1-D) function to seek the only unknown constant that is used for haze removal. Using the determined constant, the scene albedo can be recovered, eliminating the haze from single hazy images. The proposed GDM is able to suppress the halo/blocking artifacts in the recovered results due to the scene albedo, which is less sensitive to the determined constant. Simulation results on different types of benchmark images verify that the proposed technique outperforms state-of-the-art methods in terms of both recovery, quality, and real-time performance.

In this paper, a home Internet-of-Things system is analyzed by dividing it into four layers, i.e., the node layer, gateway layer, service layer, and open layer. The gateway layer, which supports a variety of wireless technologies and is the core of home wireless networks access unit, together with the node layer constitutes the home wireless network. A gateway prototype following the proposed architecture has been implemented. A testbed of an interference-aware wireless network which includes the gateway prototype has also been created for testing its user interaction performances. The experimental results show that both Wi-Fi and Bluetooth have an impact on the ZigBee communication. Considering the complex scene of home and building, ZigBee multihop communications are set to reduce the packet loss probability. In addition, an event-level-based transmission control strategy is proposed, in which the packet loss probability of wireless network is reduced by controlling the transmission priority of different levels of monitoring events, and optimizing the ZigBee wireless network channel occupancy.

Under foggy or hazy weather conditions, the visibility and color fidelity of outdoor images are prone to degradation. Hazy images can be the cause of serious errors in many computer vision systems. Consequently, image haze removal has practical significance for real-world applications. In this study, we first analyze the inherent weaknesses of the atmospheric scattering model and propose an improvement to address those weaknesses. Then, we present a fast image haze removal algorithm based on the improved model. In our proposed method, the input image is partitioned into several scenes based on the haze thickness. Next, averaging and erosion operations calculate the rough scene luminance map in a scene-wise manner. We obtain the rough scene transmission map by maximizing the contrast in each scene and then develop a way to gently remove the haze using an adaptive method for adjusting scene transmission based on scene features. In addition, we propose a guided total variation model for edge optimization, so as to prevent from the block effect as well as to eliminate the negative effect from the wrong scene segmentation results. The experimental results demonstrate that our method is effective in solving a series of common problems, including uneven illuminance, overenhanced and oversaturated images, and so forth. Moreover, our method outperforms most current dehazing algorithms in terms of visual effects, universality, and processing speed.

An important issue of supporting multi-user video streaming over wireless networks is how to optimize the systematic scheduling by intelligently utilizing the available network resources while, at the same time, to meet each video's QoS (quality of service) requirement. In this work, we study the problem of video scheduling over multi-channel multi-radio multi-hop networks with the goals of minimizing the video distortion. At first, we construct a general distortion model according to the network's transmission mechanism, as well as video's rate-distortion characteristics. Then, by joint considering the channel assignment, rate allocation and routing, we develop a fully distributed scheduling scheme to get an optimal QoS performance. Furthermore, the realization of the distributed scheduling scheme through cooperation among the channel, link and source is the highlight of this paper. Extensive simulation results are provided which demonstrate the effectiveness of our proposed scheme.