论文总字数:25143字
摘 要
在移动互联网普及的今天,智能手机已经不仅仅是一个打电话的工具,而是为我们的便捷生活做出了巨大的贡献。例如,通过智能手机可以满足支付、社交、购物等需求。智能手机在为我们提供便捷的同时,也会带来一些安全隐患。虚拟身份欺骗已成为社交媒体环境中日益重要的问题,攻击者通过脚本实现账户自动登陆、点击欺骗等恶意行为。现有的检测方案一般采用设备环境参数检测方法,但很容易被攻击者通过虚拟参数设置方式绕过。
本文通过采集智能手机加速度、陀螺仪、方向等传感器信号数据,利用机器学习算法实现用户行为识别,从而判断用户身份的真实性,以有效提高虚拟身份欺骗检测的准确性。本文主要的工作内容如下:
- 数据类别:本文采集了两类数据进行机器学习:1、真实用户数据:本文将手持手机进行输入产生的数据模拟为真实用户的输入。2、脚本数据:本文将手机置于桌面输入产生的数据模拟为脚本输入产生的数据。
- 针对传感器数据采集:本文采用了一种基于Web和NodeJS的传感器数据采集 方法。本方案可通过Web运行程序提取传感器数据并传输到后台服务器的 MongoDB数据库中。避免了不同平台采样率不同引发的误差。
- 传感器数据的预处理:针对数据奇点问题,本文采用归一化进行降噪处理。针对由采样频率以及按键时间引起的数据点数的问题,本文采用三次样条插值法进行处理。
- 数据特征处理:本文提取了数据的统计特征。
- 机器学习:本文基于Libsvm工具包对上述两种数据进行分类,从而验证通过手机传感器进行用户识别的可能性。
关键词:智能手机,传感器,机器学习,用户识别
Abstract
Today, with the popularity of mobile Internet, smart phones are not only a tool for making calls, but also a great contribution to our convenient life. As is known to all, smartphone can make the social much more convenient. However, smartphones maty result into risk because of its strength. Many of us focus on virtual identity spoofing because if its risk. Attackers use scripts to implement malicious actions such as automatic account login and click fraud. The existing detection scheme generally adopts the device environment parameter detection method, but it is easily bypassed by the attacker through the virtual parameter setting method.
This paper collects sensor signal data such as smartphone acceleration, gyroscope and direction, and uses machine learning algorithm to realize user behavior recognition, in order to judge the authenticity of user identity, which can help effectively improving the accuracy of virtual identity spoofing detection. The main work of this paper is as follows:
- Data category: This paper collects two types of data for machine learning: 1. Real user data: This paper simulates the data generated by the input of the handheld mobile phone as the input of the real user. 2, script data: This article puts the data generated by the mobile phone on the desktop input to simulate the data generated by the script input.
- For sensor data acquisition: This paper uses a Web and NodeJS based sensor data acquisition method. This solution can extract sensor data through the web running program and transfer it to the MongoDB database of the background server. Avoid errors caused by different sampling rates on different platforms.
- Preprocessing of sensor data: For the data singularity problem, this paper uses normalization for noise reduction. For the problem of the number of data points caused by the sampling frequency and the key time, this paper uses the cubic spline interpolation method for processing.
- Data feature processing: In this paper,we only get statistical characteristics.
- Machine learning: This paper classifies the above two types of data based on the Libsvm toolkit to verify the possibility of user identification through mobile phone sensors.
Keywords: smartphone, sensor, machine learning, user identification
目 录
摘要 i
Abstract ii
1 绪论 1
1.1背景 1
1.2研究现状 1
1.3论文研究内容 2
1.4论文组织结构 3
2 相关技术研究 4
2.1智能终端权限介绍 4
2.2.1加速度传感器 5
2.2.2陀螺仪传感器 5
2.23方向传感器 5
2.3机器学习 6
2.3.1机器学习简介 6
2.3.2机器学习流程 6
2.3.3分类算法 6
2.4本章小结 8
3 用户识别系统框架 9
3.1整体架构设计 9
3.2基于web传感器数据采集 11
3.2.1数据采集 11
3.2.2数据预处理 12
3.3时域特征选择和提取 12
3.4基于Libsvm的分类模型搭建与训练 14
3.5本章小结 18
4 用户识别系统的实现 19
4.1两类数据采集方案 19
4.2系统设计 19
4.3传感器数据采集系统 20
4.3.1跨平台和数据上传 20
4.3.2系统架构与工作流程 21
4.3.3 Sensor Logger 实现 22
4.3.4 Sensor Sensor 实现 24
4.4机器学习实现 25
4.4.1 数据预处理实现 25
4.4.2 基于Libsvm的机器学习实现 27
4.5实验结果与分析 28
4.5本章小结 29
5 总结与展望 31
5.1总结 31
5.2展望 31
参考文献 33
致 谢 34
1 绪论
剩余内容已隐藏,请支付后下载全文,论文总字数:25143字
该课题毕业论文、开题报告、外文翻译、程序设计、图纸设计等资料可联系客服协助查找;