基于机器视觉的无线传感器网络唤醒机制

发布时间：2018-02-20 18:56

本文关键词： 唤醒机制传感器网络哈希算法 LSH图像检索赋值权重　出处：《哈尔滨工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：近些年来,随着信息技术的不断进步,基于图像的无线传感器网络应用技术成为研究热点。在图像、视频等大数据传输中,传输数据所付出的能量代价将大幅提高,传统休眠调度机制所减少的能量消耗将不再变得有效,如何保证传感器最大限度减少传输消耗又能够同时满足有价值图像数据及时有效的传输至SINK节点成为本文的研究重点,本文以重点区域监测为应用背景,研究无线传感器网络应用中基于图像检索技术的自主唤醒机制。传统的休眠办法是人为地对传感器网络进行休眠干预以达到减少传输消耗的目的,检索唤醒的办法是由传感器网络本身对数据进行自主判断,对采集图像进行筛选,减少数据量从而达到降低能耗的目的。本文提出的基于图像检索的传感器唤醒机制提供了一种新的技术应用,将深度学习与机器视觉技术应用于无线传感器网络,主要工作内容如下:首先,研究了哈希算法,提出传感器节点初步唤醒机制,节点采集的数据经处理后转变为哈希序列,将其与背景图像哈希序列进行相似度匹配来初步判定是否有目标出现,判定目标出现后,传感器节点唤醒通信传输模块,将哈希指纹序列传输至SINK节点。其次,研究了图像差分技术以及阈值分割技术,完成传感器节点采集的图像数据中对目标区域与背景的分割,通过3DMAX软件模拟目标各姿态和场景,完成对目标不同姿态和场景下的图像库的建立,研究图像特征提取技术,建立基于目标图像库及Cifar-10图像库的GIST特征库。第三,研究图像索引技术,重点研究LSH图像检索算法,明确评价标准,在此基础上通过训练,提出基于赋值权重的WLSH检索训练算法,提高目标识别精度。最后,完成对以上三部分内容的仿真分析。
[Abstract]:In recent years, with the development of information technology, the application technology of image-based wireless sensor network (WSNs) has become a research hotspot. In the transmission of images, video and other big data, the energy cost of transmitting data will be greatly increased. The energy consumption reduced by the traditional sleep scheduling mechanism will no longer become effective. How to ensure that the sensor minimizes transmission consumption and can simultaneously meet the needs of timely and effective transmission of valuable image data to the SINK node becomes the focus of this paper. The background of this paper is the monitoring of key areas. This paper studies the automatic wake-up mechanism based on image retrieval technology in wireless sensor network application. The traditional sleep method is to intervene in sensor network sleep artificially in order to reduce transmission consumption. The method of retrieving wake-up is for the sensor network itself to judge the data independently and screen the collected images. In this paper, the sensor wake-up mechanism based on image retrieval provides a new technology application, which applies depth learning and machine vision technology to wireless sensor networks. The main work is as follows: firstly, the hashing algorithm is studied, and the initial wake-up mechanism of sensor node is proposed. The data collected by the sensor node is transformed into a hash sequence after processing. Matching the similarity with the background image hash sequence to determine whether the target appears or not, the sensor node wake up the communication transmission module, and transmit the hash fingerprint sequence to the SINK node. Secondly, after the target appears, the sensor node awakens the communication transmission module, and transmits the hash fingerprint sequence to the SINK node. The image difference technology and threshold segmentation technology are studied to complete the segmentation of the target region and background in the image data collected by sensor nodes, and the 3D Max software is used to simulate each pose and scene of the target. Complete the establishment of image database under different pose and scene of target, study the technology of image feature extraction, establish the GIST signature database based on target image database and Cifar-10 image library. Thirdly, study the image index technology, and focus on the LSH image retrieval algorithm. On the basis of training, the WLSH retrieval training algorithm based on assignment weight is proposed to improve the accuracy of target recognition. Finally, the simulation analysis of the above three parts is completed.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP212.9;TN929.5;TP391.41

【参考文献】