Convolutional neural networks for time series classification

Time series classification is an important task in time series data mining, and has attracted great interests and tremendous efforts during last decades. However, it remains a challenging problem due to the nature of time series data: high dimensionality, large in data size and updating continuously. The deep learning techniques are explored to improve the performance of traditional feature-based approaches. Specifically, a novel convolutional neural network (CNN) framework is proposed for time series classification. Different from other feature-based classification approaches, CNN can discover and extract the suitable internal structure to generate deep features of the input time series automatically by using convolution and pooling operations. Two groups of experiments are conducted on simulated data sets and eight groups of experiments are conducted on real-world data sets from different application domains. The final experimental results show that the proposed method outperforms state-of-the-art methods for time series classification in terms of the classification accuracy and noise tolerance. © 1990-2011 Beijing Institute of Aerospace Information.

Rectification of magnetic force tracker using neural network inaugmented reality system

Nonlinear errors always exist in data obtained from tracker in augmented reality （AR）, which badly influence the effect of AR. This paper proposes to rectify the errors using BP neural network. As BP neural network is prone to getting into local extrema and convergence is slow, genetic algorithm is employed to optimize the initial weights and threshold of neural network. This paper discusses how to set the crucial parameters in the algorithm. Experimental results show that the method ensures that the neural network achieves global convergence quickly and correctly. Tracking precision of AR system is improved after the tracker is rectified, and the third dimension of AR system is enhanced.

Body Worn Sensors for Health Gaming and e-Learning in Virtual Reality

Virtual reality is an emerging field in the whole world.The problem faced by people today is that they are more indulged in indoor technology rather than outdoor activities.Hence,the proposed system introduces a fitness solution connecting virtual reality with a gaming interface so that an individual can play first-person games.The system proposed in this paper is an efficient and cost-effective solution that can entertain people along with playing outdoor games such as badminton and cricket while sitting in the room.To track the human movement,sensors Micro Processor Unit(MPU6050)are used that are connected with Bluetoothmodules andArduino responsible for sending the sensor data to the game.Further,the sensor data is sent to a machine learning model,which detects the game played by the user.The detected game will be operated on human gestures.A publicly available dataset named IM-Sporting Behaviors is initially used,which utilizes triaxial accelerometers attached to the subject’s wrist,knee,and below neck regions to capture important aspects of human motion.The main objective is that the person is enjoying while playing the game and simultaneously is engaged in some kind of sporting activity.The proposed system uses artificial neural networks classifier giving an accuracy of 88.9%.The proposed system should apply to many systems such as construction,education,offices and the educational sector.Extensive experimentation proved the validity of the proposed system.

Synergy Model of Artificial Intelligence and Augmented Reality in the Processes of Exploitation of Energy Systems

The article presents a fragment of research and development, which objective was to develop technical tools and methodology to improve exploitation processes of energy systems. The author＇s model includes synergy of artificial intelligence and augmented reality. This solution, which combines modem technologies in order to improve the activities related to the continuity of energy supply, and reduce costs associated with the time needed to carry out exploitation activities and employment of qualified staff, is presented. This paper presents both theoretical foundations as well as the development of technical systems. The characteristics of exploitation processes of energy systems and possible technical conditions, as well as factors characterizing them, are discussed. The physical and software structures of the system and individual modules, as well as dependencies connecting them are demonstrated. The dependencies between physical and logical elements during the exploitation processes of energy systems, that determine decisions related to the evaluation of technical states and related activities are described. The advantages and limitations of the developed model which connects methods of data processing and analysis, interactive visualization processes and possible areas of application are as well discussed in detailed.

AN EXPLOITATION OF THE EXPERT SYSTEM OF MECHANICAL DESIGN BASED ON VIRTUAL REALITY

Object oriented methodology is introduced to build the model on expert system of mechanical design. The combination of knowledge representation with object oriented method and neural network make the expression of the complex knowledge in mechanical design world reasonably and properly, and the complicated system is built successfully. Also, the merging of virtual reality art and conventional mechanical design theory puts visual three dimensional mechanical movement into realities.

虚拟现实中基于轻量级神经网络的手势识别

手势是虚拟现实环境中的重要输入方式,手势识别的精准度和效率为虚拟现实环境的有效交互提供了基础。为了提高虚拟现实环境中手势识别的正确率和速度,提出了一种基于轻量级优化卷积神经网络的手势识别模型。首先,对手势图像样本进行特征提取及初始化,并构建卷积神经网络手势识别模型。然后,借助MobileNet V2进行轻量化处理,通过深度与逐点卷积,有效降低手势识别模型参数量和计算量。最后,根据最小化手势识别误差反向求解神经网络的各个参数。试验结果表明,所提轻量级神经网络的手势识别准确率相对于原始卷积神经网络有微小下降,但其识别效率却大幅提升,特别是在大规模手势样本的识别场景下。轻量级卷积神经网络表现出了较高的识别综合性能,在虚拟现实环境下的手势识别适应度较高。

VR环境下基于多观测角度人眼成像特性的注视估计研究

在虚拟现实(VR)的沉浸式场景中,基于先进的注视估计技术实现精确的注视点渲染,能够优化计算资源分配效率、缓解用户体验过程中可能产生的眩晕感。目前,可用的VR环境下的注视估计数据集只有单观测角度眼睛图像,缺乏不同观测角度的眼部图像数据集。文中构建了一个包含23 040张多观测角度眼睛图像的注视估计数据集与一个包含15 824张带有瞳孔标注的多观测角度眼睛图像的瞳孔检测数据集,并提出了一种结合多观测角度眼睛图像特征以相互补偿的多分支网络模型。将注视估计数据集样本用于模型的训练过程,预测欧氏距离损失可以达到7.68像素。进一步,将包含瞳孔位置信息的权重地图与图像融合,瞳孔位置信息的融合输入增强了模型的性能,欧氏距离损失降低到7.45像素。这项研究表明,所开发的模型能够提升VR环境下的注视估计精度,从而推动注视估计技术在VR产品中的广泛应用。

人体模型建模中全局映射智能估计人体形状的方法

针对人体模型建模中形状估计存在的性能低、适用性不强等问题,提出一种利用卷积神经网络进行人体形状估计的智能方法,并利用卷积神经网络层次化特征提取和多视图融合的方法精确估计人体形状。以人体模型为基础,对人体三维形状进行尺寸估计,实现快速人体形状估计,且使估计过程自动化,提高估计结果的精确性。研究表明,该方法克服了传统人工测量技术无法捕捉人体特征、估计人体形状与姿态的局限性,可用于人体模型尺寸的测量、服装虚拟现实中人体模型的建模等。

面向船舶的基于WPA-BP与虚拟现实的机舱安全监测

为了提高对船舶机舱安全运行远程故障诊断的效率,对船舶机舱的智能化诊断与监测进行了研究,设计了一种船舶机舱安全监测系统;采用优化后的狼群算法对反向传播神经网络进行改进,并创新建立了狼群算法-反向传播基础上的机舱柴油机故障诊断模型;经实验测试实现了平均诊断准确率高达99.102%、响应时间0.048 ms的验证结果;经实际应用满足了船舶安全运行的需求,提高了故障诊断有效性。

AR智能采集装置在电力设备数据检测中的应用

电力设备数据检测的过程复杂、耗时久、图像虚假,因此研究AR智能采集装置在电力设备数据检测中的应用。AR(augmented reality)智能采集装置利用头戴式设备模块的CCD(charge coupled device)数码相机获取原始电力设备AR图像,通过AR图像增强模块的空间保色法增强采集到的AR图像,结合二维小波和孪生神经网络模型,完成电力设备数据检测,并提交给数据库进行存储,同时将检测结果呈递给远程专家分析中心。实验表明应用该方法后检测过程简单,耗时少,图像增强效果佳,可高精准实现电力设备数据检测,具备较好应用效果。

虚拟现实技术辅助机器车视觉循迹建模

针对机器车真实行驶环境复杂、循迹模型训练成本较高的问题,本文提出了一种基于虚拟现实技术辅助机器车视觉循迹建模的方法。首先,通过虚拟现实技术来构建机器车虚拟行驶环境,并采集该环境中的图像数据以创建仿真数据集。然后,基于虚拟小车在虚拟行驶环境中的运行表现和循迹模型的预测准确率等指标,优化循迹模型中的网络结构以提高模型性能。最后,实现循迹模型在虚拟环境中的实时可视化循迹效果。实验结果表明,相对于传统模型和模型训练方法,改进的模型在仿真环境数据上的循迹准确率提升了5.2%,在真实环境数据集上的循迹准确率提升了5.9%,并达到了81.2%。因此,虚拟现实技术有效辅助了机器车来建立更高性能的循迹模型。

虚拟环境的系统设计方法及计算模型研究

面向对象技术和面向Ａｇｅｎｔ技术是虚拟环境系统的基本设计方法．本文用面向对象的方法来构造Ａｇｅｎｔ，并提供一组支撑Ａｇｅｎｔ的底层计算模型，如神经网络、遗传算法、专家系统和规划管理等．通过飞船游戏的模拟，详细论述了Ａｇｅｎｔ与底层计算模型之间的关系以及Ａｇｅｎｔ与环境之间的交互．本文思想和类库可应用到各种虚拟环境系统和智能应用系统．

3D交互输入新技术——手势识别

本文从分析人机交互研究的目的出发 ,介绍了用户产生手势到系统“感知”手势的过程 ,包括手势的建模及识别的原理和方法 .本文在对手结构和手关节运动分析的基础上 ,提出了基于数据手套和基于视觉(摄象机 )的手势输入方式 ,通过手势图象的分析 ,采用模糊神经网络方法对手势图象进行识别 ,取得了较好的效果 ,实现了一种在虚拟环境中具有广阔运用前景的

手模型分析及手势识别

对手结构和关节运动作了深人地分析，并建立了对其施加约束的手模型。根据所定义的约束及运动类型，建立了适合于手势合成的手数据结构。并采用神经网络理论进行手势识别，取得了较好的效果。

基于神经网络的手势识别

本文在对手结构和关节运动作了深入分析的基础上，建立了对其施加约束的手模型．根据所定义的约束及运动类型，建立了适合于手势合成的手数据结构．并采用神经网络理论进行手势识别，取得了较好的效果．

增强现实系统中磁力跟踪器的校正

提出了利用BP神经网络对跟踪器进行校正。针对神经网络训练速度慢、容易陷入局部极值的情况,首先利用具有良好全局搜索能力的遗传算法来优化BP神经网络的各层初始权值和阈值,为后续神经网络的搜索定位出一个优化的搜索空间。实验结果表明,利用该遗传神经网络方法进行跟踪器校正,能够显著提高增强现实系统的精度,有助于提高增强现实系统的真实感。

基于模糊神经网络的手势识别

本文通过对人手结构和手关节运动的分析 ,建立了手模型及其相应的数据结构 .由数据手套的数据接口获取各指节的曲伸角度 ,建立手势标准样本库 ,并提出了采用模糊神经网络进行手势识别 ,用手势标准样本加以训练 ,使其具备识别手势的功能 .试验结果表明 。

虚拟现实技术在数字展示设计中的应用研究

在信息技术飞速发展的时代,电子计算机被应用到了各行各业,也带动了整个社会的发展。虚拟现实技术正是这一时代的产物。本文以神经网络知识为基础理论,针对虚拟现实技术在数字展示设计中的仿真性和可观性进行评价。同时,通过大量的问卷调查,采集了相关的数据,建立了关于虚拟现实技术在数字展示设计中的应用评价模型。将黄山、泰山、故宫和乌镇作为研究对象,进行了模型的验证。评价结果显示,黄山和故宫的数字展示效果良好,泰山和乌镇的展示效果相对较差,这一结论与实际情况相符。从而,说明了模型的合理性。

用ANN模型扩展COMGIS系统的地震灾害模拟仿真功能

利用组件式GIS(COMGIS)技术来开发结合了专业震害分析模型的城市地震灾害模拟仿真系统,在VB下调用Matlab的神经网络工具箱来扩展GIS的地震灾害模拟仿真功能。研究表明:设定地震可随机输入,在系统菜单下直接调用水平成层土地震反应分析程序SHAKE91来计算出与设定震级相对应的场地土层地震动参数;实现了基于BP神经网络的震害模拟仿真模型在COMGIS系统中的模块化;系统的GIS空间分析功能可使模拟仿真结果与空间信息相匹配,实现了随机假定地震下城市潜在地震灾害的快速评估。

一种交互输入新技术——三维手势识别

结合文献资料及笔者的科研成果 ,综合介绍了基于数据手套和基于视觉 (摄象机 )输入手势的新的三维交互输入技术。这种技术利用计算机进行图象分析 ,并采用神经网络和几何方法对输入三维图象进行识别 ,从而实现三维的交互输出。这是一种具有广阔运用前景的新技术。