基于双重注意力孪生网络的实时视觉跟踪

为了解决全卷积孪生网络（SiamFC）跟踪算法在跟踪目标经历剧烈的外观变化时容易发生模型漂移从而导致跟踪失败的问题，提出了一种双重注意力机制孪生网络（DASiam）去调整网络模型并且不需要在线更新。首先，主干网络使用修改后表达能力更强的并适用于目标跟踪任务的VGG网络；然后，在网络的中间层加入一个新的双重注意力机制去动态地提取特征，这种机制由通道注意机制和空间注意机制组成，分别对特征图的通道维度和空间维度进行变换得到双重注意特征图；最后，通过融合两个注意机制的特征图进一步提升模型的表征能力。在三个具有挑战性的跟踪基准库即OTB2013、OTB100和2017年视觉目标跟踪库（VOT2017）实时挑战上进行实验，实验结果表明，以40 frame/s的速度运行时，所提算法在OTB2013和OTB100上的成功率指标比基准SiamFC分别高出3.5个百分点和3个百分点，并且在VOT2017实时挑战上面超过了2017年的冠军SiamFC，验证了所提出算法的有效性。     

基于深度卷积长短时神经网络的视频帧预测

针对视频帧预测中难以准确预测空间结构信息细节的问题，通过对卷积长短时记忆（LSTM）神经网络的改进，提出了一种深度卷积长短时神经网络的方法。首先，将输入序列图像输入到两个不同通道的深度卷积LSTM网络组成的编码网络中，由编码网络学习输入序列图像的位置信息变化特征和空间结构信息变化特征；然后，将学习到的变化特征输入到与编码网络通道数对应的解码网络中，由解码网络输出预测的下一张图；最后，将这张图输入回解码网络中，预测接下来的一张图，循环预先设定的次后输出全部的预测图。与卷积LSTM神经网络相比，在Moving-MNIST数据集上的实验中，相同训练步数下所提方法不仅保留了位置信息预测准确的特点，而且空间结构信息细节表征能力更强。同时，将卷积门控循环单元（GRU）神经网络的卷积层加深后，该方法在空间结构信息细节表征上也取得了提升，检验了该方法思想的通用性。     

Deep reinforcement learning-based cooperative interactions among heterogeneous vehicular networks

Most real-world vehicle nodes can be structured into an interconnected network of vehicles. Through structuring these services and vehicle device interactions into multiple types, such internet of vehicles becomes multidimensional heterogeneous overlay networks. The heterogeneousness of the overlays makes it difficult for the overlay networks to coordinate with each other to improve their performance. Therefore, it poses an interesting but critical challenge to the effective analysis of heterogeneous virtual vehicular networks. A variety of virtual vehicular networks can be easily deployed onto the native network by applying the concept of SDN (Software Defined Networking). These virtual networks reflect their heterogeneousness due to their different performance goals, and they compete for the same physical resources of the underlying network, so that a sub-optimal performance of the virtual networks may be achieved. Therefore, we propose a Deep Reinforcement Learning (DRL) approach to make the virtual networks cooperate with each other through the SDN controller. A cooperative solution based on the asymmetric Nash bargaining is proposed for co-existing virtual networks to improve their performance. Moreover, the Markov Chain model and DRL resolution are introduced to leverage the heterogeneous performance goals of virtual networks. The implementation of the approach is introduced, and simulation results confirm the performance improvement of the latency sensitive, loss-rate sensitive and throughput sensitive heterogeneous vehicular networks using our cooperative solution.     

改进HSR-FCN的服装图像识别分类算法研究

目前网络上的服装图像数量增长迅猛，对于大量服装图像实现智能分类的需求日益增加。将基于区域的全卷积网络（Region-Based Fully Convolutional Networks，R-FCN）引入到服装图像识别中，针对服装图像分类中网络训练时间长、形变服装图像识别率低的问题，提出一种新颖的改进框架HSR-FCN。新框架将R-FCN中的区域建议网络和HyperNet网络相融合，改变图片特征学习方式，使得HSR-FCN可以在更短的训练时间内达到更高的准确率。在模型中引入了空间转换网络，对输入服装图像和特征图进行了空间变换及对齐，加强了对多角度服装和形变服装的特征学习。实验结果表明，改进后的HSR-FCN模型有效地加强了对形变服装图像的学习，且在训练时间更短的情况下，比原来的网络模型R-FCN平均准确率提高了大约3个百分点，达到96.69%。     

Distributed adaptive neural networks leader-following formation control for quadrotors with directed switching topologies

The leader-following formation problem is discussed for a team of quadrotors under directed switching topologies. To obtain a more general dynamic model, we describe the quadrotor system in a non-affine pure-feedback form with mismatched unknown nonlinearities. By employing an adaptive neural networks state observer to approximate the unknown nonlinear functions and to reconstruct the immeasurable inner states, we propose a novel distributed output feedback formation control protocol with the backstepping method combining with the dynamic surface control technique. From the Lyapunov stability theorem, all signals in the closed-loop formation system are proven to be cooperatively semiglobally uniformly ultimately bounded for any given bounded initial conditions. Finally, we proved that we verify the performance of the proposed formation control approach by a simulation study.     

An efficient numerical scheme for solving fractional infinite-horizon optimal control problems

This paper presents an efficient numerical method to solve fractional infinite-horizon optimal control problems, where the dynamic control system depends on Caputo fractional derivatives. First, by a suitable change of variable, we transform the fractional infinite-horizon optimal control problem to a finite-horizon one. Then, with the help of an approximation, we replace the Caputo derivative to integer order derivative. According to the Pontryagin minimum principle (PMP) for optimal control problems and by constructing an error function, we define an unconstrained minimization problem. In the optimization problem, we use trial solutions for state, costate and control functions where these trial solutions are constructed by using two-layered perceptron neural network. Some numerical results are introduced to explain our main results.     

基于概率感知模型的多节点联合网络覆盖算法

针对传感网在对关注目标节点进行覆盖过程中出现的非连续性覆盖，将会产生覆盖盲区以及大量冗余数据迫使通信链路出现瓶颈现象，提出了一种基于概率感知模型的多节点联合覆盖算法（Multi-Node Joint Network Coverage Algorithm，MNJC）。该算法借助于概率相关知识，给出了多节点联合时覆盖质量期望值计算方法，以减少覆盖盲区的产生，提高了多节点联合网络覆盖率；利用节点动态转换机制对节点部署进行重新优化，抵制了网络能量的快速消耗，延长网络生存周期；该算法与其他三种算法覆盖率、生存周期以及节点存活数量等方面进行了比对实验，其性能指标平均提升了18.11%，15.57%和8.26%，从而验证该算法的有效性和实效性。     

基于DRGAN和支持向量机的合成孔径雷达图像目标识别

为解决SAR图像目标识别中样本缺乏和方位角敏感问题,提出了一种基于DRGAN和SVM的SAR图像目标识别算法。首先,采用多尺度分形特征对SAR图像进行增强,经过分割得到目标二值图像,基于Hu二阶矩估计目标的方位角。然后对估计得到的目标方位角进行量化编码,结合原始图像作为输入,对设计的DRGAN模型参数进行训练与优化。由于DRGAN中的深度生成模型将目标姿态与外观表示进行解耦设计,故可利用该模型将SAR图像样本变换到同一方位角区间。基于变换后的训练样本分别提取归一化灰度特征,利用SVM训练分类器。采用MSTAR数据集在多个不同操作条件下对提出的算法进行测试,实验结果表明,在带变体的标准操作条件下,能够达到97.97%的分类精度,优于部分基于CNN模型的分类精度,在4种扩展操作条件下的分类精度分别为97.83%,91.77%,97.11%和97.04%,均优于传统方法的分类精度。在SAR图像目标方位角估计存在一定误差的情况下,训练得到的GAN模型作为SAR图像目标旋转估计器,能够使得在不进行复杂样本预处理的前提下,仍然取得较高的SAR图像目标识别精度。     

Secure lightweight password authenticated key exchange for heterogeneous wireless sensor networks

Several three-party password authenticated key exchange (3-PAKE) protocols have recently been proposed for heterogeneous wireless sensor networks (HWSN). These are efficient and designed to address security concerns in ad-hoc sensor network applications for a global Internet of Things framework, where a user may request access to sensitive information collected by resource-constrained sensors in clusters managed by gateway nodes. In this paper we first analyze three recently proposed 3-PAKE protocols and discuss their vulnerabilities. Then, based on Radio Frequency Identification technologies we propose a novel 3-PAKE protocol for HWSN applications, with two extensions for additional security features, that is provably secure, efficient and flexible.     

Predicting pollutant removal in constructed wetlands using artificial neural networks(ANNs)

Growth in urban population,urbanisation,and economic development has increased the demand for water,especially in water-scarce regions.Therefore,sustainable approaches to water management are needed to cope with the effects of the urbanisation on the water environment.This study aimed to design novel configurations of tidal-flow vertical subsurface flow constructed wetlands(VFCWs) for treating urban stormwater.A series of laboratory experiments were conducted with semi-synthetic influent stormwater to examine the effects of the design and operation variables on the performance of the VFCWs and to identify optimal design and operational strategies,as well as maintenance requirements.The results show that the VFCWs can significantly reduce pollutants in urban stormwater,and that pollutant removal was related to specific VFCW designs.Models based on the artificial neural network(ANN) method were built using inputs derived from data exploratory techniques,such as analysis of variance(ANOVA) and principal component analysis(PCA).It was found that PCA reduced the dimensionality of input variables obtained from different experimental design conditions.The results show a satisfactory generalisation for predicting nitrogen and phosphorus removal with fewer variable inputs,indicating that monitoring costs and time can be reduced.