计算机工程与应用 ›› 2017, Vol. 53 ›› Issue (21): 98-102.DOI: 10.3778/j.issn.1002-8331.1605-0261

• 网络、通信与安全 • 上一篇    下一篇

基于时域特征提取的围栏入侵模式分类方法

周  静1,赵鲁阳1,罗炬锋1,2   

  1. 1.中国科学院 上海微系统与信息技术研究所 无线传感网与通信重点实验室,上海 200050
    2.上海物联网有限公司,上海 200050
  • 出版日期:2017-11-01 发布日期:2017-11-15

Fence intrusion pattern classification method based on time domain feature extraction

ZHOU Jing1, ZHAO Luyang1, LUO Jufeng1,2   

  1. 1.Key Laboratory of Wireless Sensor Network & Communication, Shanghai Institute of Microsystem and Information Technology, Chinese Academy of Sciences, Shanghai 200050, China
    2.Shanghai Internet of Things Co., Ltd., Shanghai 200050, China
  • Online:2017-11-01 Published:2017-11-15

摘要: 基于无线传感网的防入侵应用领域中行为分类问题,提出一种基于时域特征提取的电子围栏入侵检测及异常入侵模式识别系统。由于频域处理方法计算量大、复杂度高、传感器采样率高,为减轻系统的传输负担并减少时延,首先对原始信号预处理提取时域特征,然后通过一个三层的BP神经网络对目标事件进行分类,最后对比了多种典型的分类器方法的准确率。仿真结果表明:相比于频域处理方法,该方法复杂度低、易于实现,多种分类器准确率达86%以上,其中BP神经网络测试集的准确率能够达到94%,并且训练集和测试集的准确率偏差较小。

关键词: 无线传感网, 时域特征提取, 围栏入侵, BP神经网络, 模式分类

Abstract: Focused on the issue of behavior classification in the field of security application based on wireless sensor networks, an electronic fence intrusion detection and abnormal pattern classification system is proposed using time domain feature extraction. The method of frequency?domain’s feature extraction contains massive computation with expensive complexity, and the sensors’ sampling rate is high. In order to reduce the system’s transmission burden and time delay, firstly, the raw data is preprocessed to extract time domain features. Then a three-layer BP neural networks classifier is used to classify the target events. Lastly, the accuracy rate of several kinds of typical classifiers are compared. Simulation results indicate that, compared with the method of feature extraction in frequency domain, this method is low in complexity and easy to implement, and the accuracy rate can reach more than 86%. What’s more, for the BP neural networks, the accuracy deviation between the training and testing set is relatively small, while the accuracy is reaching 94% for the testing data set which is higher than others.

Key words: wireless sensor networks, time domain feature extraction, fence intrusion, BP neural networks, pattern classification