【作者】 刘莹；

【导师】 黄玉妍；

【作者基本信息】 哈尔滨师范大学 ， 计算机技术， 2021， 硕士

【摘要】 近年来,群智感知成为计算领域的前沿研究问题,群智感知通过参与者所携带的智能手机和智能手机所独有的计算和感知能力进行采集感知数据。应用到周围环境的监控、道路路况的监控和智能交通等相关领域。通过移动用户所携带的移动设备,对范围较大的感知任务的感知数据进行采集,相对于其他技术这样可以达到高效率的完成感知任务。参与者完成感知任务的同时,感知任务的下发和感知信息的上传在群智感知网络中是被广泛的研究和关注,感知数据的分类能够更加精准的为数据需求者提供数据是尤为重要的,因为目前的研究中感知数据分类很少,不能更加精准的为数据需求者提供相应数据,对感知数据的分类就需要研究人员进一步研究。感知数据的分类问题已经被广泛的研究与关注,本文提出了一种基于胶囊网络的群智感知数据分类模型。首先,对群智感知国内外现状进行分析,对群智感知面临的问题、主要特征以及数据分类等相关内容进行详细的介绍。对群智感知系统模型的群智感知平台、移动用户、数据分类模型、感知任务、数据请求者以及群智感知任务的定义进行详细描述,为之后建立群智感知系统模型奠定理论基础,然后,群智感知平台感知数据类型搭建胶囊网络的结构,构建了感知数据分类模型的样本数据,引入基于胶囊网络的感知数据分类模型实现对样本数据特征的学习,并结合动态路由算法和损失函数实现了卷积层-初级胶囊-数字胶囊参数的优化。最后,对试验环境、胶囊网络的结构和参数传递可视化进行描述,对基于胶囊网络的数据分类模型进行模型参数分析、模型优化器分析、模型准确度与误差分析。仿真实验表明基于胶囊网络的群智感知数据分类模型相比基于残差网络的群智感知数据分类模型,感知数据的分类准确率更高,更能准确的为请求者提供数据。更多还原

【Abstract】 In recent years,crowd sensing has become a cutting-edge research problem in the field of computing.crowd sensing collects perceptual data through the smart phones carried by participants and the unique computing and perception capabilities of smart phones.It is applied to the surrounding environment monitoring,road condition monitoring and intelligent transportation and other related fields.Through the mobile devices carried by mobile users,the perception data of a wide range of sensing tasks can be collected.Compared with other technologies,the sensing tasks can be completed with high efficiency.Participants complete perception tasks at the same time,the sense of mission of the upload issued and perception of information in the crowd sensing network is widely research and attention,the classification of the perception data can be more accurate to provide data to the data of demanders is particularly important,because the basic no sensory data classification in the current study,it will not be able to more accurate to provide corresponding data to data demanders,the categorization of perception data requires researchers to further study.The classification of perception data has been widely studied and concerned.In this paper,we propose a classification model of crowd sensing perception data based on capsule network.Firstly,the status quo of crowd sensing at home and abroad is analyzed,and the problems,main characteristics and data classification of crowd sensing are introduced in detail.Group of crowd sensing of crowd sensing system model platform,mobile users and data classification model,the sense of mission,data requesters and crowd sensing tasks are described in detail,the definition of crowd sensing system model is established to lay theoretical basis,then,crowd sensing of platform perception data type structures,capsule network structure,the constructed model of perception data classification of sample data,the introduction of the sensory data classification model based on capsule network to study the features of sample data,and combining with dynamic routing algorithm and the loss function convolution layer-primary capsule-a digital capsule parameters optimization.Finally,the experimental environment,the structure of capsule network and the visualization of parameter transmission are described,and the model parameter analysis,model optimizer analysis,model accuracy and error analysis are carried out for the data classification model based on capsule network.The simulation experiment shows that compared with the residual network based crowd sensing data classification model,the classification accuracy of the perception data is higher and it can provide data for the requester more accurately.更多还原