结合时空注意力机制和自适应图卷积网络的骨架行为识别

针对骨架行为识别对时空特征提取不充分以及难以捕捉全局上下文信息的问题,研究了一种将时空注意力机制和自适应图卷积网络相结合的人体骨架行为识别方案.首先,构建基于非局部操作的时空注意力模块,辅助模型关注骨架序列中最具判别性的帧和区域;其次,利用高斯嵌入函数和轻量级卷积神经网络的特征学习能力,并考虑人体先验知识在不同时期的影...     

基于三维图卷积与注意力增强的行为识别模型

针对当前行为识别方法无法有效提取非欧式3维骨架序列的时空信息与缺乏针对特定关节关注的问题,该文提出了一种基于3维图卷积与注意力增强的行为识别模型.首先,介绍了3维卷积与图卷积的具体工作原理;其次,基于图卷积中可处理变长邻居节点的图卷积核,引入3维卷积的3维采样空间将2维图卷积核改进为具有3维采样空间的3维图卷积核,提出...     

基于时空特征增强图卷积网络的骨架行为识别

针对骨架行为识别不能充分挖掘时空特征的问题,该文提出一种基于时空特征增强的图卷积行为识别模型(STFE-GCN)。首先,介绍表征人体拓扑结构邻接矩阵的定义及双流自适应图卷积网络模型的结构,其次,采用空域上的图注意力机制,根据邻居节点的重要性程度分配不同的权重系数,生成可充分挖掘空域结构特征的注意力系数矩阵,并结合非局部网络生成的全局邻接矩阵,提出一种新的空域自适应邻接矩阵,以期增强对人体空域结构特征的提取;然后,时域上采用混合池化模型以提取时域关键动作特征和全局上下文特征,并结合时域卷积提取的特征,以期增强对行为信息中时域特征的提取。再者,在模型中引入改进通道注意力网络(ECA-Net)进行通道注意力增强,更有利于模型提取样本的时空特征,同时结合空域特征增强、时域特征增强和通道注意力,构建时空特征增强图卷积网络模型在多流网络下实现端到端的训练,以期实现时空特征的充分挖掘。最后,在NTU-RGB+D和NTU-RGB+D120两个大型数据集上开展骨架行为识别研究,实验结果表明该模型具有优秀的识别准确率和泛化能力,也进一步验证了该模型充分挖掘时空特征的有效性。     

基于骨架动作识别的协作卷积Transformer网络

近年来,基于骨架的人体动作识别任务因骨架数据的鲁棒性和泛化能力而受到了广泛关注。其中,将人体骨骼建模为时空图的图卷积网络取得了显著的性能。然而图卷积主要通过一系列3D卷积来学习长期交互联系,这种联系偏向于局部并且受到卷积核大小的限制,无法有效地捕获远程依赖关系。该文提出一种协作卷积Transformer网络(Co-ConvT),通过引入Transformer中的自注意力机制建立远程依赖关系,并将其与图卷积神经网络(GCNs)相结合进行动作识别,使模型既能通过图卷积神经网络提取局部信息,也能通过Transformer捕获丰富的远程依赖项。另外,Transformer的自注意力机制在像素级进行计算,因此产生了极大的计算代价,该模型通过将整个网络分为两个阶段,第1阶段使用纯卷积来提取浅层空间特征,第2阶段使用所提出的ConvT块捕获高层语义信息,降低了计算复杂度。此外,原始Transformer中的线性嵌入被替换为卷积嵌入,获得局部空间信息增强,并由此去除了原始模型中的位置编码,使模型更轻量。在两个大规模权威数据集NTU-RGB+D和Kinetics-Skeleton上进行实验验证,该模型分别达到了88.1%和36.6%的Top-1精度。实验结果表明,该模型的性能有了很大的提高。     

基于深度自编码-高斯混合模型的视频异常检测方法

由于异常定义的模糊性和真实数据的复杂性,视频异常检测是智能视频监控中最具挑战性的问题之一。基于自动编码器(AE)的帧重建（当前或未来帧）是一种流行的视频异常检测方法。使用在正常数据上训练的模型,异常场景的重建误差通常比正常场景的重建误差大得多。但是,这类方法忽略了正常数据本身的内部结构,效率较低。基于此,提出了一种深度自动编码高斯混合模型(DAGMM)。首先利用深度自动编码器获得输入视频片段的生成低维表示和重构误差,并将其进一步输入高斯混合模型(GMM)。而估计网络则通过高斯混合模型预测能量概率,然后通过能量密度概率判断异常。DAGMM以端到端的方式同时联合优化深度自动编码器和GMM的参数,能够平衡自动编码重建、低维表示的密度估计和正则化,泛化能力强。在两个公共基准数据集上的实验结果表明,DAGMM达到了现有最高技术发展水平,在UCSD Ped2和ShanghaiTech两个数据集上分别取得了95.7%和72.9%的帧级AUC。     

基于贝叶斯融合的时空流异常行为检测模型

针对直接利用卷积自编码网络未考虑视频时间信息的问题,该文提出基于贝叶斯融合的时空流异常行为检测模型。空间流模型采用卷积自编码网络对视频单帧进行重构,时间流模型采用卷积长短期记忆(LSTM)编码-解码网络对短期光流序列进行重构。接着,分别计算空间流模型和时间流模型下每帧的重构误差,设计自适应阈值对重构误差图进行二值化,并基于贝叶斯准则对空间流和时间流下的重构误差进行融合,得到融合重构误差图,并在此基础上进行异常行为判断。实验结果表明,该算法在UCSD和Avenue视频库上的检测效果优于现有异常检测算法。

     

A comparative study between single and multi-frame anomaly detection and localization in recorded video streams

Video anomaly detection is usually studied by considering the spatial and temporal contexts. This paper focuses first on spatial context and shows that it can be a fast real-time solution. In the first part of this work there are two main contributions: employing a new deep network for reconstruction and introducing a new regularity scoring function. The new deep architecture is based on pyramid of input images and compared to UNet, the proposed architecture boosts AUC by 15% and the new regularity scoring function is based on SSIM. The second part employs a multiframe approach to distinguish temporal behavior anomalies. The second approach enhances the results by 7% compared to spatial anomaly detection. Comparing the two approaches, if computing power is limited and real time anomaly detection is looked for, single frame detection is preferred while multi frame analysis offers a much wider possibility of anomaly detection.     

基于多尺度偶数卷积注意力U-Net的医学图像分割

为了提高U-Net网络性能的同时尽可能减少额外计算量,本文提出了一种新的多尺度偶数卷积注意力UNet(Multiscale Even Convolution Attention U-Net,MECAU-Net)网络。该网络在编码端采用2×2偶数卷积代替3×3卷积进行特征提取,并借鉴多尺度思想,采用4×4偶数卷积将得到的信息直接传递给主干部分,以获取更全面的图像信息并减少额外计算开销,同时还采用对称填充解决偶数卷积提取信息过程中产生的偏移问题。此外,在2×2偶数卷积模块后加入卷积注意力模块,结合空间和通道注意力,在提取更丰富的信息的同时几乎不增加额外开销。最后,在两个医学图像数据集上进行仿真实验,实验结果表明提出的MECAU-Net网络相对于U-Net在稍微增加计算成本的情况下,分割性能得到了较大的提升,并比其他对比网络取得更好的分割性能的同时还降低了参数量。     

基于Word Embedding的遥感影像检测分割

遥感影像检测分割技术通常需提取影像特征并通过深度学习算法挖掘影像的深层特征来实现.然而传统特征（如颜色特征、纹理特征、空间关系特征等）不能充分描述影像语义信息,而单一结构或串联算法无法充分挖掘影像的深层特征和上下文语义信息.针对上述问题,本文通过词嵌入将空间关系特征映射成实数密集向量,与颜色、纹理特征的结合.其次,本文构建基于注意力机制下图卷积网络和独立循环神经网络的遥感影像检测分割并联算法（Attention Graph Convolution Networks and Independently Recurrent Neural Network,ATGIR）.该算法首先通过注意力机制对结合后的特征进行概率权重分配;然后利用图卷积网络（GCNs）算法对高权重的特征进一步挖掘并生成方向标签,同时使用独立循环神经网络（IndRNN）算法挖掘影像特征中的上下文信息,最后用Sigmoid分类器完成影像检测分割任务.以胡杨林遥感影像检测分割任务为例,我们验证了提出的特征提取方法和ATGIR算法能有效提升胡杨林检测分割任务的性能.     

基于体素化图卷积网络的三维点云目标检测方法

针对激光雷达点云的稀疏性和空间离散分布的特点,通过结合体素划分和图表示方法设计了新的图卷积特征提取模块,提出一种基于体素化图卷积神经网络的激光雷达三维点云目标检测算法。该方法通过消除传统3D卷积神经网络的计算冗余性,不仅提升了网络的目标检测能力,并且提高了点云拓扑信息的分析能力。文中设计的方法在KITTI公开数据集的车辆、行人、骑行者的3D目标检测和鸟瞰图目标检测任务的检测性能相比基准网络均有了有效提升,尤其在车辆3D目标检测任务上最高提升了13.75%。实验表明:该方法采用图卷积特征提取模块有效提高了网络整体检测性能和数据拓扑关系的学习能力,为三维点云目标检测任务提供了新的方法。     

基于改进SSD的人群异常行为检测算法研究

针对人群异常行为检测任务中存在的算法复杂度较高,重叠遮挡等带来的检测精度低等问题,本文提出一种基于改进SSD（Single Shot Multi-box Detector）的人群异常行为检测算法。首先采用轻量级网络MobileNet v2代替原始特征提取网络VGG-16,并通过可变形卷积模块构建卷积层来增强感受野,然后通过将位置信息整合到通道注意力中来进行特征增强,能够捕获空间位置之间的远程依赖关系,从而可以较好处理重叠遮挡问题。实验结果表明,本文提出的算法对人群异常行为具有较好的检测效果。     

A multiscale hypothesis testing approach to anomaly detection andlocalization from noisy tomographic data

In this paper, we investigate the problems of anomaly detection and localization from noisy tomographic data. These are characteristic of a class of problems that cannot be optimally solved because they involve hypothesis testing over hypothesis spaces with extremely large cardinality. Our multiscale hypothesis testing approach addresses the key issues associated with this class of problems. A multiscale hypothesis test is a hierarchical sequence of composite hypothesis tests that discards large portions of the hypothesis space with minimal computational burden and zooms in on the likely true hypothesis. For the anomaly detection and localization problems, hypothesis zooming corresponds to spatial zooming - anomalies are successively localized to finer and finer spatial scales. The key challenges we address include how to hierarchically divide a large hypothesis space and how to process the data at each stage of the hierarchy to decide which parts of the hypothesis space deserve more attention. For the latter, we pose and solve a nonlinear optimization problem for a decision statistic that maximally disambiguates composite hypotheses. With no more computational complexity, our optimized statistic shows substantial improvement over conventional approaches. We provide examples that demonstrate this and quantify how much performance is sacrificed by the use of a suboptimal method as compared to that achievable if the optimal approach were computationally feasible.     

Iterative graph filtering network for 3D human pose estimation

Graph convolutional networks (GCNs) have proven to be an effective approach for 3D human pose estimation. By naturally modeling the skeleton structure of the human body as a graph, GCNs are able to capture the spatial relationships between joints and learn an efficient representation of the underlying pose. However, most GCN-based methods use a shared weight matrix, making it challenging to accurately capture the different and complex relationships between joints. In this paper, we introduce an iterative graph filtering framework for 3D human pose estimation, which aims to predict the 3D joint positions given a set of 2D joint locations in images. Our approach builds upon the idea of iteratively solving graph filtering with Laplacian regularization via the Gauss–Seidel iterative method. Motivated by this iterative solution, we design a Gauss–Seidel network (GS-Net) architecture, which makes use of weight and adjacency modulation, skip connection, and a pure convolutional block with layer normalization. Adjacency modulation facilitates the learning of edges that go beyond the inherent connections of body joints, resulting in an adjusted graph structure that reflects the human skeleton, while skip connections help maintain crucial information from the input layer’s initial features as the network depth increases. We evaluate our proposed model on two standard benchmark datasets, and compare it with a comprehensive set of strong baseline methods for 3D human pose estimation. Our experimental results demonstrate that our approach outperforms the baseline methods on both datasets, achieving state-of-the-art performance. Furthermore, we conduct ablation studies to analyze the contributions of different components of our model architecture and show that the skip connection and adjacency modulation help improve the model performance.     

Homology analysis of malware based on graph

Malware detection and homology analysis has been the hotspot of malware analysis.API call graph of malware can represent the behavior of it.Because of the subgraph isomorphism algorithm has high complexity,the analysis of malware based on the graph structure with low efficiency.Therefore,this studies a homology analysis method of API graph of malware that use convolutional neural network.By selecting the key nodes,and construct neighborhood receptive field,the convolution neural network can handle graph structure data.Experimental results on 8 real-world malware family,shows that the accuracy rate of homology malware analysis achieves 93%,and the accuracy rate of the detection of malicious code to 96%.     

基于多任务学习和注意力机制的双分支深度换脸检测方法

深度换脸技术的出现严重威胁了公众的隐私安全。为了解决现有深度换脸检测方法的局限性,基于多任务学习策略提出了一种双分支检测网络,实现在检测视频伪造的同时逐帧检测。该网络引入了注意力机制和时序学习模块,通过学习局部空间信息和时序信息提升检测性能。该方法在公开数据集Celeb-DF和FaceForensics++上获得了比当前先进换脸检测方法更高的准确率和ROC(Receiver Operating Characteristic)曲线下面积（Area under ROC Curve,AUC）,面对不同光照、人脸朝向、视频质量时表现出了良好的鲁棒性。     

GAD:topology-aware time series anomaly detection

To solve the problems of anomaly detection,intelligent operation,root cause analysis of node equipment in the network,a graph-based gated convolutional codec anomaly detection model was proposed for time series data such as link delay,network throughput,and device memory usage.Considering the real-time requirements of network scenarios and the impact of network topology connections on time series data,the time dimension features of time series were extracted in parallel based on gated convolution and the spatial dependencies were mined through graph convolution.After the encoder composed of the spatio-temporal feature extraction module encoded the original input time series data,the decoder composed of the convolution module was used to reconstruct the time series data.The residuals between the original data and the reconstructed data were further used to calculate the anomaly score and detect anomalies.Experiments on public data and simulation platforms show that the proposed model has higher recognition accuracy than the current time series anomaly detection benchmark algorithm.     

Global relational reasoning with spatial temporal graph interaction networks for skeleton-based action recognition

With the prevalence of accessible depth sensors, dynamic skeletons have attracted much attention as a robust modality for action recognition. Convolutional neural networks (CNNs) excel at modeling local relations within local receptive fields and are typically inefficient at capturing global relations. In this article, we first view the dynamic skeletons as a spatio-temporal graph (STG) and then learn the localized correlated features that generate the embedded nodes of the STG by message passing. To better extract global relational information, a novel model called spatial–temporal graph interaction networks (STG-INs) is proposed, which perform long-range temporal modeling of human body parts. In this model, human body parts are mapped to an interaction space where graph-based reasoning can be efficiently implemented via a graph convolutional network (GCN). After reasoning, global relation-aware features are distributed back to the embedded nodes of the STG. To evaluate our model, we conduct extensive experiments on three large-scale datasets. The experimental results demonstrate the effectiveness of our proposed model, which achieves the state-of-the-art performance.     

融合知识图谱与图片特征的推荐模型

目前知识图谱研究主要面向信息检索、自然语言理解等领域,在推荐系统中融合知识图谱成为推荐领域学者广泛关注的问题。为了解决单一知识图谱忽略的丰富知识信息,该文对知识图谱进行多模态扩展,并提出一种融合知识图谱与图片特征的推荐模型(KG-I)。不同于其他基于知识图谱的推荐算法,该方法增加视觉嵌入、知识嵌入和结构嵌入去挖掘用户项目之间的隐式反馈信息。该模型利用深度游走模型(Deep Walk)捕获空间结构的方法和波纹网络模型(RippleNet)挖掘知识图谱的知识表达的思想,并且考虑图片对用户偏好的影响,有效地将信息进行融合,并在真实数据集上与其他模型实验比较,研究多种特征的影响,分析不同稀疏度数据下的表现。结果表明,融合知识图谱与图片特征的个性化推荐模型完全优于其他的对比算法并且有效缓解数据稀疏情况。     

Saliency detection in human crowd images of different density levels using attention mechanism

The human visual system has the ability to rapidly identify and redirect attention to important visual information in high complexity scenes such as the human crowd. Saliency prediction in the human crowd scene is the process using computer vision techniques to imitate the human visual system, predicting which areas in a human crowd scene may attract human attention. However, it is a challenging task to identify which factors may attract human attention due to the high complexity of the human crowd scene. In this work, we propose Multiscale DenseNet — Dilated and Attention (MSDense-DAt), a convolutional neural network (CNN) using self-attention to integrate the result of knowledge-driven gaze in the human visual system to identify salient areas in the human crowd scene. Our method combines various state-of-the-art deep learning architectures to deal with the high complexity in human crowd image, such as multiscale DenseNet for multiscale deep features extraction, self-attention, and dilated convolution. Then the effectiveness of each component in our CNN architecture is evaluated by comparing different components combinations. Finally, the proposed method is further evaluated in different crowd density levels to appraise the effect of crowd density on model performance.