In this paper,a new architecture of optical networks—the optical network based on server system is considered.From the point of this new architecture,the network can be modeled as a server system with three type serv...In this paper,a new architecture of optical networks—the optical network based on server system is considered.From the point of this new architecture,the network can be modeled as a server system with three type servers—the access server,the node server and the link server.The network performances such as cost,energy consume and network capacity can be affected by the capability of these three type servers.New ILP formulations are proposed to analyze the network capacity under two types of node severs,with and without wavelength converter.Computer simulations are conducted to evaluate the effectiveness of these new formulations.The study has shown that the network can achieve the same throughput under the two types of node servers and the network throughput increases when the maximum allowed variation increases.展开更多
Considering wireless sensor network characteristics,this paper uses network simulator,version2(NS-2)algorithm to improve Ad hoc on-demand distance vector(AODV)routing algorithm,so that it can be applied to wireless se...Considering wireless sensor network characteristics,this paper uses network simulator,version2(NS-2)algorithm to improve Ad hoc on-demand distance vector(AODV)routing algorithm,so that it can be applied to wireless sensor networks.After studying AODV routing protocol,a new algorithm called Must is brought up.This paper introduces the background and algorithm theory of Must,and discusses the details about how to implement Must algorithm.At last,using network simulator(NS-2),the performance of Must is evaluated and compared with that of AODV.Simulation results show that the network using Must algorithm has perfect performance.展开更多
Integrating the power grid technology with renewable power generation technologies, Demand Response (DR) programs enabled by the Advanced Metering Infrastructure (AMI) were introduced into the power grid in the intere...Integrating the power grid technology with renewable power generation technologies, Demand Response (DR) programs enabled by the Advanced Metering Infrastructure (AMI) were introduced into the power grid in the interest of both utilities and residents. They help to achieve load balance and increase the grid reliability by encouraging residents to reduce their power usage during peak load periods in return for incentives. To automate this process, appliances, in-house sensors, and the AMI controller need to be networked together. In this paper, we compare mainstream network technologies applicable to home appliance control and propose a solution combining Power Line Communication (PLC) with wireless communication in smart homes for the purpose of energy saving. We extended NS-2, a popular network simulator, to model such combined network scenarios. Using a number of different routing strategies, we then model and evaluate the network performance of DR programs in smart homes in such a combined network.展开更多
针对肺炎X光图像病灶区域特征难以被提取和现有模型轻量化程度不高的问题,提出一种特征融合的MV2-Transformer(FFMV2-Transformer)肺炎X光图像分类模型。首先,采用轻量型网络MobileNetV2(Mobile Network Version 2)作为主干网络,并在反...针对肺炎X光图像病灶区域特征难以被提取和现有模型轻量化程度不高的问题,提出一种特征融合的MV2-Transformer(FFMV2-Transformer)肺炎X光图像分类模型。首先,采用轻量型网络MobileNetV2(Mobile Network Version 2)作为主干网络,并在反向残差瓶颈块中嵌入坐标注意力(CA)机制,从而通过将位置信息嵌入通道信息提高模型对病灶区域特征的提取能力;其次,设计局部和全局特征融合模块(LGFFM)将卷积层提取的局部特征与Transformer捕获的全局特征相结合,从而使模型能同时捕捉病灶区域的细节信息和整体信息,并进一步提高模型的语义特征提取能力;最后,设计跨层特征融合模块(CFFM)将空间注意力机制增强的浅层特征的空间信息与通道注意力机制增强的深层特征的语义信息相结合,从而获得丰富的上下文信息。为了验证模型的有效性,在肺炎X光数据集上进行消融实验和对比实验,结果表明,FFMV2-Transformer模型与MobileViT(Mobile Vision Transformer)模型相比,准确率、精确率、召回率、F1值和AUC(Area Under ROC(Receiver Operating Characteristic)Curve)值分别提高了1.09、0.31、1.91、1.08和0.40个百分点。可见,FFMV2-Transformer模型能在实现模型轻量化的同时,有效提取肺炎X光图像病灶区域的特征。展开更多
为推动输电线路智能化巡检模式,本文针对人机协同巡检模式下的图像差异大及干扰因素多等问题,提出一种轻量化改进型YOLOv8(you only look once version 8)的多类别绝缘子缺陷检测算法。首先在特征提取网络中融合可变形大核注意力的同时...为推动输电线路智能化巡检模式,本文针对人机协同巡检模式下的图像差异大及干扰因素多等问题,提出一种轻量化改进型YOLOv8(you only look once version 8)的多类别绝缘子缺陷检测算法。首先在特征提取网络中融合可变形大核注意力的同时进行轻量化,提升网络对不同目标轮廓与尺寸的适用性;其次引入渐进的特征融合策略以改善不同层次特征间的语义差距,提高网络的检测精度;并设计轻量化非对称检测头,进一步减少参数冗余;最后改进边框损失函数有效降低由密集遮挡造成的漏检和误检数量。实验结果表明,本文算法相较于原算法检测精度提升了7.7%,参数量和计算量分别减少了26.4%和30.2%,并在密集、遮挡、多类别目标缺陷检测中的评价指标均领先于当前主流的几类目标检测算法,显著提高了复杂环境下的多类别绝缘子缺陷检测,实现了检测精度和速度的双重提升。展开更多
随着网络技术的持续演进,为满足日益增长的通信需求,急需对油气自控网络进行升级。基于虚拟交换实例(Virtual Switching Instance,VSI)/虚拟扩展局域网(Virtual eXtensible Local Area Network,VXLAN)隧道通信技术,开展网际互连协议第4...随着网络技术的持续演进,为满足日益增长的通信需求,急需对油气自控网络进行升级。基于虚拟交换实例(Virtual Switching Instance,VSI)/虚拟扩展局域网(Virtual eXtensible Local Area Network,VXLAN)隧道通信技术,开展网际互连协议第4版/网际互连协议第6版(Internet Protocol Version 4/Internet Protocol Version 6,IPv6)双栈部署与效能评估相关研究。通过确立双栈架构的设计原则与目标,构建分层网络架构,实现VSI/VXLAN隧道与IPv4/IPv6双栈架构的深度融合,完成核心网络设备与接入层网络的双栈部署,并实施安全域划分。基于多维度效能评估指标体系,采集多测点、多负载条件下的性能数据,量化评估网络的安全防护能力。实验结果表明,该方案可显著提升油气自控网络在协议兼容性、传输效率与安全性方面的综合性能,为其数字化升级提供理论支撑与实践路径。展开更多
针对小尺度目标在检测时精确率低且易出现漏检和误检等问题,提出一种改进的YOLOv3(You Only Look Once version 3)小目标检测算法。在网络结构方面,为提高基础网络的特征提取能力,使用DenseNet-121密集连接网络替换原Darknet-53网络作...针对小尺度目标在检测时精确率低且易出现漏检和误检等问题,提出一种改进的YOLOv3(You Only Look Once version 3)小目标检测算法。在网络结构方面,为提高基础网络的特征提取能力,使用DenseNet-121密集连接网络替换原Darknet-53网络作为其基础网络,同时修改卷积核尺寸,进一步降低特征图信息的损耗,并且为增强检测模型对小尺度目标的鲁棒性,额外增加第4个尺寸为104×104像素的特征检测层;在对特征图融合操作方面,使用双线性插值法进行上采样操作代替原最近邻插值法上采样操作,解决大部分检测算法中存在的特征严重损失问题;在损失函数方面,使用广义交并比(GIoU)代替交并比(IoU)来计算边界框的损失值,同时引入Focal Loss焦点损失函数作为边界框的置信度损失函数。实验结果表明,改进算法在VisDrone2019数据集上的均值平均精度(mAP)为63.3%,较原始YOLOv3检测模型提高了13.2百分点,并且在GTX 1080 Ti设备上可实现52帧/s的检测速度,对小目标有着较好的检测性能。展开更多
基金supported by China Post-doctoral Science Foundation funded project(20070420013)Open Fund of National Laboratory on Local Fiber-Optic Communication Networks&Advanced optical Communication Systems,(Pe-king University),PRChinaGuangxi Science Foundation(0731003)
文摘In this paper,a new architecture of optical networks—the optical network based on server system is considered.From the point of this new architecture,the network can be modeled as a server system with three type servers—the access server,the node server and the link server.The network performances such as cost,energy consume and network capacity can be affected by the capability of these three type servers.New ILP formulations are proposed to analyze the network capacity under two types of node severs,with and without wavelength converter.Computer simulations are conducted to evaluate the effectiveness of these new formulations.The study has shown that the network can achieve the same throughput under the two types of node servers and the network throughput increases when the maximum allowed variation increases.
文摘Considering wireless sensor network characteristics,this paper uses network simulator,version2(NS-2)algorithm to improve Ad hoc on-demand distance vector(AODV)routing algorithm,so that it can be applied to wireless sensor networks.After studying AODV routing protocol,a new algorithm called Must is brought up.This paper introduces the background and algorithm theory of Must,and discusses the details about how to implement Must algorithm.At last,using network simulator(NS-2),the performance of Must is evaluated and compared with that of AODV.Simulation results show that the network using Must algorithm has perfect performance.
文摘Integrating the power grid technology with renewable power generation technologies, Demand Response (DR) programs enabled by the Advanced Metering Infrastructure (AMI) were introduced into the power grid in the interest of both utilities and residents. They help to achieve load balance and increase the grid reliability by encouraging residents to reduce their power usage during peak load periods in return for incentives. To automate this process, appliances, in-house sensors, and the AMI controller need to be networked together. In this paper, we compare mainstream network technologies applicable to home appliance control and propose a solution combining Power Line Communication (PLC) with wireless communication in smart homes for the purpose of energy saving. We extended NS-2, a popular network simulator, to model such combined network scenarios. Using a number of different routing strategies, we then model and evaluate the network performance of DR programs in smart homes in such a combined network.
文摘针对肺炎X光图像病灶区域特征难以被提取和现有模型轻量化程度不高的问题,提出一种特征融合的MV2-Transformer(FFMV2-Transformer)肺炎X光图像分类模型。首先,采用轻量型网络MobileNetV2(Mobile Network Version 2)作为主干网络,并在反向残差瓶颈块中嵌入坐标注意力(CA)机制,从而通过将位置信息嵌入通道信息提高模型对病灶区域特征的提取能力;其次,设计局部和全局特征融合模块(LGFFM)将卷积层提取的局部特征与Transformer捕获的全局特征相结合,从而使模型能同时捕捉病灶区域的细节信息和整体信息,并进一步提高模型的语义特征提取能力;最后,设计跨层特征融合模块(CFFM)将空间注意力机制增强的浅层特征的空间信息与通道注意力机制增强的深层特征的语义信息相结合,从而获得丰富的上下文信息。为了验证模型的有效性,在肺炎X光数据集上进行消融实验和对比实验,结果表明,FFMV2-Transformer模型与MobileViT(Mobile Vision Transformer)模型相比,准确率、精确率、召回率、F1值和AUC(Area Under ROC(Receiver Operating Characteristic)Curve)值分别提高了1.09、0.31、1.91、1.08和0.40个百分点。可见,FFMV2-Transformer模型能在实现模型轻量化的同时,有效提取肺炎X光图像病灶区域的特征。
文摘为推动输电线路智能化巡检模式,本文针对人机协同巡检模式下的图像差异大及干扰因素多等问题,提出一种轻量化改进型YOLOv8(you only look once version 8)的多类别绝缘子缺陷检测算法。首先在特征提取网络中融合可变形大核注意力的同时进行轻量化,提升网络对不同目标轮廓与尺寸的适用性;其次引入渐进的特征融合策略以改善不同层次特征间的语义差距,提高网络的检测精度;并设计轻量化非对称检测头,进一步减少参数冗余;最后改进边框损失函数有效降低由密集遮挡造成的漏检和误检数量。实验结果表明,本文算法相较于原算法检测精度提升了7.7%,参数量和计算量分别减少了26.4%和30.2%,并在密集、遮挡、多类别目标缺陷检测中的评价指标均领先于当前主流的几类目标检测算法,显著提高了复杂环境下的多类别绝缘子缺陷检测,实现了检测精度和速度的双重提升。
文摘随着网络技术的持续演进,为满足日益增长的通信需求,急需对油气自控网络进行升级。基于虚拟交换实例(Virtual Switching Instance,VSI)/虚拟扩展局域网(Virtual eXtensible Local Area Network,VXLAN)隧道通信技术,开展网际互连协议第4版/网际互连协议第6版(Internet Protocol Version 4/Internet Protocol Version 6,IPv6)双栈部署与效能评估相关研究。通过确立双栈架构的设计原则与目标,构建分层网络架构,实现VSI/VXLAN隧道与IPv4/IPv6双栈架构的深度融合,完成核心网络设备与接入层网络的双栈部署,并实施安全域划分。基于多维度效能评估指标体系,采集多测点、多负载条件下的性能数据,量化评估网络的安全防护能力。实验结果表明,该方案可显著提升油气自控网络在协议兼容性、传输效率与安全性方面的综合性能,为其数字化升级提供理论支撑与实践路径。
文摘针对小尺度目标在检测时精确率低且易出现漏检和误检等问题,提出一种改进的YOLOv3(You Only Look Once version 3)小目标检测算法。在网络结构方面,为提高基础网络的特征提取能力,使用DenseNet-121密集连接网络替换原Darknet-53网络作为其基础网络,同时修改卷积核尺寸,进一步降低特征图信息的损耗,并且为增强检测模型对小尺度目标的鲁棒性,额外增加第4个尺寸为104×104像素的特征检测层;在对特征图融合操作方面,使用双线性插值法进行上采样操作代替原最近邻插值法上采样操作,解决大部分检测算法中存在的特征严重损失问题;在损失函数方面,使用广义交并比(GIoU)代替交并比(IoU)来计算边界框的损失值,同时引入Focal Loss焦点损失函数作为边界框的置信度损失函数。实验结果表明,改进算法在VisDrone2019数据集上的均值平均精度(mAP)为63.3%,较原始YOLOv3检测模型提高了13.2百分点,并且在GTX 1080 Ti设备上可实现52帧/s的检测速度,对小目标有着较好的检测性能。
