Delay-dependent exponential stability for uncertain neutral stochastic neural networks with interval time-varying delay

This paper is mainly concerned with the problem for the robustly exponential stability in mean square moment of uncertain neutral stochastic neural networks with interval time-varying delay. With an appropriate augmented Lyapunov–Krasovskii functional (LKF) formulated, the convex combination method is utilised to estimate the derivative of the LKF. Some new delay-dependent exponential stability criteria for such systems are obtained in terms of linear matrix inequalities, which involve fewer matrix variables and have less conservatism. Finally, two illustrative numerical examples are given to show the effectiveness of our obtained results.     

超像素条件随机场下的RGB-D视频显著性检测

目的 视觉显著性在众多视觉驱动的应用中具有重要作用,这些应用领域出现了从2维视觉到3维视觉的转换,从而基于RGB-D数据的显著性模型引起了广泛关注。与2维图像的显著性不同,RGB-D显著性包含了许多不同模态的线索。多模态线索之间存在互补和竞争关系,如何有效地利用和融合这些线索仍是一个挑战。传统的融合模型很难充分利用多模态线索之间的优势,因此研究了RGB-D显著性形成过程中多模态线索融合的问题。方法 提出了一种基于超像素下条件随机场的RGB-D显著性检测模型。提取不同模态的显著性线索,包括平面线索、深度线索和运动线索等。以超像素为单位建立条件随机场模型,联合多模态线索的影响和图像邻域显著值平滑约束,设计了一个全局能量函数作为模型的优化目标,刻画了多模态线索之间的相互作用机制。其中,多模态线索在能量函数中的权重因子由卷积神经网络学习得到。结果 实验在两个公开的RGB-D视频显著性数据集上与6种显著性检测方法进行了比较,所提模型在所有相关数据集和评价指标上都优于当前最先进的模型。相比于第2高的指标,所提模型的AUC（area under curve）,sAUC（shuffled AUC）,SIM（similarity）,PCC（Pearson correlation coefficient）和NSS（normalized scanpath saliency）指标在IRCCyN数据集上分别提升了2.3%,2.3%,18.9%,21.6%和56.2%;在DML-iTrack-3D数据集上分别提升了2.0%,1.4%,29.1%,10.6%,23.3%。此外还进行了模型内部的比较,验证了所提融合方法优于其他传统融合方法。结论 本文提出的RGB-D显著性检测模型中的条件随机场和卷积神经网络充分利用了不同模态线索的优势,将它们有效融合,提升了显著性检测模型的性能,能在视觉驱动的应用领域发挥一定作用。     

Delay-dependent robust stability analysis for interval linear time-variant systems with delays and application to delayed neural networks

Some delay-dependent robust asymptotical stability criteria for uncertain linear time-variant systems with multiple delays are established by means of parameterized first-order model transformation and the transformation of the interval uncertainty into the norm-bounded uncertainty. The stable regions with respect to the delay parameters are also formulated. Based on these results, we investigate the stability issue of a class of delayed neural networks that can be transformed into linear time-variant systems, and then several new global asymptotical stability criteria are exploited. Numerical examples are presented to illustrate the effectiveness of our results.     

Entropy maximisation and open queueing networks with priorities and blocking

A review is carried out on the characterisation and algorithmic implementation of an extended product-form approximation, based on the principle of maximum entropy (ME), for a wide class of arbitrary finite capacity open queueing network models (QNMs) with service and space priorities. A single server finite capacity GE/GE/1/N queue with R (R>1) distinct priority classes, compound Poisson arrival processes (CPPs) with geometrically distributed batches and generalised exponential (GE) service times is analysed via entropy maximisation, subject to suitable GE-type queueing theoretic constraints, under preemptive resume (PR) and head-of-line (HOL) scheduling rules combined with complete buffer sharing (CBS) and partial buffer sharing (PBS) management schemes stipulating a sequence of buffer thresholds {N=(N₁,…,N_R),0<N_i≤N_i−1,i=2,…,R}. The GE/GE/1/N queue is utilised, in conjunction with GE-type first two moment flow approximation formulae, as a cost-effective building block towards the establishment of a generic ME queue-by-queue decomposition algorithm for arbitrary open QNMs with space and service priorities under repetitive service blocking with random destination (RS-RD). Typical numerical results are included to illustrate the credibility of the ME algorithm against simulation for various network topologies and define experimentally pessimistic GE-type performance bounds. Remarks on the extensions of the ME algorithm to other types of blocking mechanisms, such as repetitive service blocking with fixed destination (RS-FD) and blocking-after-service (BAS), are included.     

融合自编码器和one-class SVM的异常事件检测

目的 在自动化和智能化的现代生产制造过程中,视频异常事件检测技术扮演着越来越重要的角色,但由于实际生产制造中异常事件的复杂性及无关生产背景的干扰,使其成为一项非常具有挑战性的任务。很多传统方法采用手工设计的低级特征对视频的局部区域进行特征提取,然而此特征很难同时表示运动与外观特征。此外,一些基于深度学习的视频异常事件检测方法直接通过自编码器的重构误差大小来判定测试样本是否为正常或异常事件,然而实际情况往往会出现一些原本为异常的测试样本经过自编码得到的重构误差也小于设定阈值,从而将其错误地判定为正常事件,出现异常事件漏检的情形。针对此不足,本文提出一种融合自编码器和one-class支持向量机（support vector machine,SVM）的异常事件检测模型。方法 通过高斯混合模型（Gaussian mixture model,GMM）提取固定大小的时空兴趣块（region of interest,ROI）;通过预训练的3维卷积神经网络（3D convolutional neural network,C3D）对ROI进行高层次的特征提取;利用提取的高维特征训练一个堆叠的降噪自编码器,通过比较重构误差与设定阈值的大小,将测试样本判定为正常、异常和可疑3种情况之一;对自编码器降维后的特征训练一个one-class SVM模型,用于对可疑测试样本进行二次检测,进一步排除异常事件。结果 本文对实际生产制造环境下的机器人工作场景进行实验,采用AUC （area under ROC）和等错误率（equal error rate,EER）两个常用指标进行评估。在设定合适的误差阈值时,结果显示受试者工作特征（receiver operating characteristic,ROC）曲线下AUC达到91.7%,EER为13.8%。同时,在公共数据特征集USCD （University of California,San Diego） Ped1和USCD Ped2上进行了模型评估,并与一些常用方法进行了比较,在USCD Ped1数据集中,相比于性能第2的方法,AUC在帧级别和像素级别分别提高了2.6%和22.3%;在USCD Ped2数据集中,相比于性能第2的方法,AUC在帧级别提高了6.7%,从而验证了所提检测方法的有效性与准确性。结论 本文提出的视频异常事件检测模型,结合了传统模型与深度学习模型,使视频异常事件检测结果更加准确。