The development of generative adversarial networks (GANs) has revolutionized image generation and editing. However, the capacity to create realistic images presents serious security concerns, particularly in the context of face-based payment systems. Deepfakes leverages GANs to generate manipulated videos or images, which may present opportunities for identity theft and fraudulent transactions. For instance, perpetrators employ Deepfakes technology to forge identifying information about victims, such as transplanting their faces into fake videos or images to make it appear like they are performing activities they have never done before. To address this growing concern, this study proposes a deep learning-based detection method utilizing an improved convolutional neural network model. The proposed model comprises two key modules, namely the multiscale attention (MA) module and the halo attention (HA) module. Specifically, MA is designed to recognize faces and other details in the forged image. HA is built to focus on localized regions of the image. Experimental results show that the proposed model scores 97.12 and 99.32 on FF++ (HQ) dataset and 91.26 and 95.43 on FF++ (LQ) dataset in terms of accuracy and area under curve, respectively. The remarkable accuracy and performance make it a dependable solution for safeguarding face payment systems.

Image quality assessment (IQA) methods are generally studied in the spatial or transform domain. Due to the BRISQUE algorithm evaluating the quality of an image only based on its natural scene statistics of the spatial domain, the frequency features that are extracted from the modulation transfer function (MTF) are applied to improve its performance. MTF is estimated based on the slanted-edge method. The two-dimensional grey fitting algorithm is utilised to estimate the edge slope more accurately. Then the three-order Fermi function is utilised to match the preliminary estimated edge spread function to reduce the aliasing influence on MTF estimation. The features such as crucial frequency and the MTF value at Nyquist frequency are calculated and adopted to the BRISQUE method to assess the image quality. Experimental results on the image quality assessment databases illustrated that the proposed method outperforms the BRISQUE method and some other common methods, based on the linear and nonlinear correlation between the image quality assessed by the methods and their subjective value.

With the rapid development of Internet of Everything and artificial intelligence techniques and massive amounts of video surveillance data, crowd counting has drawn extensive attention in computer vision. Inspired by deep learning methods, convolutional neural networks (CNN) have been dedicated to improving the effectiveness of crowd counting. As CNN is unable to capture the continuous size changes of heads in images, the large-scale variations impede the development of crowd counting. To solve this problem, this paper presents an attention and multi-feature fused network (AMFNet) containing a multi-level feature extractor and four attentional density estimator (ADE) modules. The multi-level extractor is used to extract the features of different sizes and various kinds of context information based on a deep network backbone. The existing ADE modules are built to merge different level features to generate a high-quality density map. A channel attention unit is adopted in the ADE modules to identify the head accurately. Then, four ADE modules are applied to exploit multi-level features and generate a fine-grained density map for coping with various scales. The experiment results show that the proposed AMFNet performs well in dense crowd scenarios, and that it is comparable to mainstream methods in terms of accuracy and robustness.

Super-resolution (SR) technology plays a crucial role in improving the spatial resolution of remote sensing images so as to overcome the physical limitations of spaceborne imaging systems. Although deep convolutional neural networks have achieved promising results, most of them overlook the advantage of self-similarity information across different scales and high-dimensional features after the upsampling layers. To address the problem, we propose a hybrid-scale hierarchical transformer network (HSTNet) to achieve faithful remote sensing image SR. Specifically, we propose a hybrid-scale feature exploitation module to leverage the internal recursive information in single and cross scales within the images. To fully leverage the high-dimensional features and enhance discrimination, we designed a cross-scale enhancement transformer to capture long-range dependencies and efficiently calculate the relevance between high-dimension and low-dimension features. The proposed HSTNet achieves the best result in PSNR and SSIM with the UCMecred dataset and AID dataset. Comparative experiments demonstrate the effectiveness of the proposed methods and prove that the HSTNet outperforms the state-of-the-art competitors both in quantitative and qualitative evaluations.

Computed tomography (CT) imaging has been widely used in clinical medicine, and high-resolution CT images play a crucial role in the determination of lesions. To fully excavate the contributive information of initial features and improve the feature representation ability of the model, we propose a pixel-attention feedback network (PAFNet) for CT image super-resolution reconstruction. Specifically, the PAFNet adopts multi-feedback network as backbone to make full use of initial features. Subsequently, a gated feedback (GF) block is introduced to refine the underlying features using the feedback features. To enrich the output characteristics and pay attention to essential details, a pixel attention mechanism is adopted to the self-calibration convolution. The subjective and objective evaluation demonstrate the superiority of the proposed method over the state-of-the-art approaches.

现有模块化多电平换流器(MMC)的调制策略主要基于比较型排序方式,存在算法复杂、计算负荷大、子模块(SM)频繁投切、系统损耗较大等缺陷.针对上述问题,文中提出一种基于动态均压偏差率(DVDR)的系统调制方式.首先,将子模块电容电压与其平均值作差得到差值序列,将差值序列按正负进行快速分组;然后,针对所得子序列设置动态均压偏差率,计算得到必要动作子模块(NASM),只需对于单一子序列中必要动作子模块序列进行排序即可降低系统计算负荷;其次,结合优化开关频率控制设置不同电压阈值与降频因子,通过判断桥臂电流方向,在保持电压一致性的前提下降低功率器件开关频率,减少电压波形畸变率;最后,在Matlab/Simulink平台上搭建仿真实验模型,验证文中所述控制策略的可行性.

Retinal fundus images play significant roles in the early detection and treatment of various ocular diseases. However, they are often suffered from low luminance in the process of shooting. To address this problem, we propose a Complexity Reduction Retinex (CR2) model for the enhancement of low luminance retinal fundus images. The proposed method enables the divided illumination component to be spatially smooth and the reflectance component to be piece-wise continuous. Meanwhile, to improve the computational efficiency, we divide the illumination and reflection components into two independent subproblems and solve them efficiently by Alternating Direction Minimizing (ADM) method. Comparative results demonstrate that the proposed method outperforms the state-of-the-art methods in terms of qualitative and quantitative evaluations.

Crowd counting is a practical yet essential research topic in computer vision, which has been beneficial to diverse applications in smart city environment safety. The commonly adopted paradigm in most existing methods is to regress a Gaussian density map that works as the learning objective during model training. However, given the unavoidable identity occlusion and scale variation in a crowd image, the corresponding Gaussian density map is degraded, failing to provide reliable supervision for optimization. To address this problem, we propose to replace the traditional Gaussian density map with a better alternation, namely the smooth inverse map (SIM). The proposed SIM can reflect the head location spatially and provide a smooth gradient to stabilize the model learning. Besides, we want the method to learn more discriminative features to cope with the challenge of large-scale variations. We deliver a multiscale aggregation (MA) to adaptively fuse features in different hierarchies to benefit semantic information under diverse receptive filed. The SIM and MA are meant to be complementary modules to guide the model in learning an accurate density map. Extensive experiments on benchmark datasets demonstrate the effectiveness of the proposed method compared with the state-of-the-art techniques.

Low-rank and sparse decomposition can be used in video surveillance analysis.This composition is more difficult to perform on Compressive Sensing(CS) videos than on videos sampled by the Nyquist theorem.Inspired by the method of projecting Nyquist-sampled signals into the orthogonal space of their low-rank section, two different CS and projection methods are proposed for the low-rank and sparse decomposition of CS videos.The first method projects measurements of CS videos into the orthogonal space of their low-rank partition.The second method performs the CS measurement after projection.For both methods, the operator can measure the CS videos and calculate the orthogonal projection simultaneously. The sparse foreground is reconstructed via CS, followed by the low-rank background. Because the progressive change in the background may change the orthogonal space of the low-rank matrices, the Structural Similarity(SSIM) metric is used to assess whether the orthogonal space of the low-rank matrices is changed.Experimental results show that compared with the SpaRcs method, the proposed method achieves better low-rank and sparse decomposition results, as well as improves the Peak Signal-to-Noise Ratio(PSNR) of the recovery images by a maximum level of 2 dB.

Images captured in a low-light environment are strongly influenced by noise and low contrast, which is detrimental to tasks such as image recognition and object detection. Retinex-based approaches have been continuously explored for low-light enhancement. Nevertheless, Retinex decomposition is a highly ill-posed problem. The estimation of the decomposed components should be combined with proper constraints. Meanwhile, the noise mixed in the low-light image causes unpleasant visual effects. To address these problems, we propose a Constraint Low-Rank Approximation Retinex model (CLAR). In this model, two exponential relative total variation constraints were imposed to ensure that the illumination is piece-wise smooth and that the reflectance component is piece-wise continuous. In addition, the low-rank prior was introduced to suppress the noise in the reflectance component. With a tailored separated alternating direction method of multipliers (ADMM) algorithm, the illumination and reflectance components were updated accurately. Experimental results on several public datasets verify the effectiveness of the proposed model subjectively and objectively.

The low-light image enhancement plays a crucial role in computer vision and multimedia applications. However, it is still a challenging task, as the degraded images reduce the visual naturalness and visibility. To address this problem, we build a novel variational Retinex model to accurately estimate the illumination and reflectance components. The illumination and reflectance are jointly updated by alternating optimization algorithm. Experimental results on several public datasets demonstrate that the proposed method outperforms the state-of-the-art methods in Retinex decomposition and illumination adjustment.

Crowd counting is a considerable yet challenging task in intelligent video surveillance and urban security systems. The performance has been significantly boosted along with the springing up of the convolutional neural networks (CNNs). However, accurate and efficient crowd counting in congested scenes remains under-explored due to scale variation and cluttered background. To address these problems, we propose a biologically inspired crowd counting method named group-split attention network (GSANet). The GSANet consists of three principal modules, namely GS module, dual-aware attention module, and aggregation module. The GS module processes the subfeatures of each group in parallel, and groups the input feature map to reduce the computational cost. The dual-aware attention module synergies the spatial and channel dimensional information to alleviate the estimation error in background regions. The aggregation module adopts a learning-based cross-group strategy to aggregate and facilitate the fusion of feature maps along different channel dimensions. Extensive experimental results on five benchmark crowd datasets demonstrate that the GSANet achieves superior performances in terms of accuracy and efficiency.

Crowd counting in congested scenes is a crucial yet challenging task in video surveillance and urban security system. The performance of crowd counting has been greatly boosted with the rapid development of deep learning. However, robust crowd counting in high-density environment with scale variations remains under-explored. To address this problem, we propose a dual attention-aware network ( DA^2 Net) for robust crowd counting in dense crowd scene with scale variations. Specifically, the DA^2 Net consists of two modules, namely Spatial Attention (SA) module and Channel Attention (CA) module. The SA module focuses on the spatial dependencies in the whole feature map to locate the heads accurately. The CA module attempts to handle the relations between channel maps and highlights the discriminative information in specific channels. Thus, it alleviates the mistaken estimation for background regions. The interactions between SA module and CA module provide the synergy which facilitates the learning of discriminative features with a focus on the essential head region. Experimental results on five benchmark datasets, i.e., ShanghaiTech, UCF_CC_50, UCF-QNRF, WorldExpo’10, and NWPU, demonstrate that the DA^2 Net can achieve the state-of-the-art performance on both accuracy and robustness.

The goal of abnormal behavior detection is to detect an anomalous event in video as accurate as possible. Motion information is crucial in such case as an inadequate motion estimation can easily make it worse. In this work, an abnormal event detection method was proposed to detect the occurrence of an anomaly automatically by using generative adversarial network (GAN) and streak flow acceleration. The proposed method is mainly composed of two components: (1) GAN-based framework that feeds on motion patterns to detect abnormal events, and (2) explicitly modeling motion information by incorporating streak flow acceleration. The effectiveness of the proposed model is verified on public benchmarks and comparative results show that our method performs favorably against many state-of-the-art methods.

模块化多电平换流器(modular multilevel converter,MMC)中子模块电容电压波动与环流相互影响,环流二倍频分量进入子模块储能电容,造成子模块电压波动,引起其他偶次谐波环流.针对该问题,提出一种拓展环流抑制器(ex tended circulating current suppressor,ECCS),将动态最速下降(dynamic steepest descent,DSD)算法与陷波器结合,对二倍频环流与其他偶次谐波环流进行精确检测.结合拓展准PR环流抑制器,利用环流选择器,对逆变侧负载发生改变时产生的突变环流设定环流抑制区间进行最优化抑制,提高了系统中环流检测的准确性与收敛速度,扩展了环流抑制器应用范围.在MATLAB/Simulink中搭建双端MMC有源输电网络,验证了所提控制策略的有效性.

针对电动汽车锂离子电池整数阶模型不能精确反映电池极化反应的问题,提出了一种基于自适应遗传算法(AGA)的分数阶模型,并采用分数阶多新息卡尔曼滤波(FOMIEKF)算法对电池荷电状态(SOC)估计.在二阶RC等效电路模型的基础上建立分数阶模型并用AGA辨识模型参数,然后用FOMIEKF算法进行SOC估计,最后与卡尔曼滤波(EKF)、分数阶扩展卡尔曼滤波(FOEKF)算法进行比较.结果表明,在混合动力脉冲测试下,模型端电压最大误差低于1％,SOC平均误差与最大误差比传统方法分别下降了0.79％、0.95％.因此,基于AGA分数阶模型的FOMIEKF方法可以有效估计SOC.

电晕放电会对高压输电线路造成极大危害,因此,检测电晕故障对于电力系统的安全具有重大意义.为了准确定位高压设备电晕放电的故障点,提出了一种基于IHS(Intensity,Hue,Saturation)和小波变换的可见光与紫外光的图像融合算法,首先对可见光进行IHS变换,将得到的I分量与紫外光进行小波分解得到各自的高低频分量,对于低频分量采取加权融合的融合算法,高频分量则采取基于区域特性的融合算法,然后通过小波重构得到新的高低频分量,最后进行IHS逆变换产生新的融合图像.实验结果表明,文中方法在可见光与紫外光融合处理中取得了较好的融合效果,优于IHS变换算法与传统的小波变换算法,且图像更加清晰,细节纹理也更加丰富,有效地保留了边缘信息.

针对电力设备红外图像存在对比度较低、噪声复杂、清晰度不优的问题,提出一种基于NSCT和改进Pal_King算法的电力设备红外图像增强方法.采用直方图双向均衡化对红外测温图像进行预处理,对NSCT变换产生的低高频子带分别采用线性增强和改进的自适应阈值函数进行系数修正,通过NSCT反变换重构图像,利用改进的Pal_King算法进行图像增强处理.实验结果表明,该方法输出图像的平均梯度值、峰值信噪比值和熵值分别保持在5.9、25.4和6.8以上,提高了电力设备红外图像的对比度,有效抑制了噪声,使图像更加清晰.

The photon counting image collected under 10− 4 lux environment has a degraded image quality due to background noises and other problems. Bayesian estimation is a classical approach for photon counting image restoration and regularization has also been widely used in image processing. However, the regularization method is not suitable for photon counting images with extremely lack of information, and the recovery effect of Bayesian estimation in images mixed with unknown noise is not ideal. The main contribution of this paper is that on the basis of Bayesian estimation, the regularization method is introduced to solve the problem of restoring photon counting images mixed with unknown noise under 10− 4 lux environment. The original part is that the gamma distribution of the expected value of photon counting is used as its prior condition, and the error function is expressed as the form of the norm to establish the objective function. Through an approximate iterative solution, the optimal estimation of the photon counting expectation is carried out to achieve the optimal restoration of the photon counting image. Experiments demonstrate that the background noise is effectively removed and the image quality is improved after restoring photon counting images. Also, the final result of the proposed method is superior to other comparative methods in multiple evaluation indexes and achieved better effects.