【作者】 常晓敏；

【导师】 赵涓涓；

【作者基本信息】 太原理工大学 ， 计算机技术， 2016， 硕士

【摘要】 森林资源是国家的重要自然资源,森林火灾的发生会给国家造成严重的损失,因此利用先进的技术手段对森林火灾进行监控,对保护森林资源不受损失非常重要。近年来,随着物联网技术理论的不断发展,其应用领域也不断扩展,森林火灾的监测和预警就是其应用之一。无线传感器网络(WSN)和无线多媒体传感器网络(WMSN)作为物联网的底层网络,在森林火灾的智能监控中,体现出其覆盖范围广,成本低,数据准确性高等特点。但是,WSN和WMSN属于资源受限的网络,其供电设备、计算能力和带宽都是有限的。而基于物联网的火灾监控预警系统中传感网络层会产生大量监测数据,数据的传输消耗大量网络资源与计算资源,严重影响传感网络使用寿命。针对WSN和WMSN中存在的问题,本文提出了有效的信息融合算法,以减少底层传感网络中数据的传输量,节省能源,提高监测和预警效率:1)基于WSN的标量传感器的数据融合针对森林火灾发生前后的特点,提出一种基于传输概率阈值的数据融合算法,将WSN中的温度、湿度、红外、烟雾等标量传感器监测值进行融合。该算法首先利用加权平均算法,计算传感器测量值在该传感器所在节点的权系数;其次利用逻辑回归模型得到该节点的火灾发生概率;再次把该节点的火灾发生概率与阈值进行比较,大于阈值则将该节点的火灾概率值向簇头节点发送,否则不发送;从而有效减少了底层网络中无效的监测数据的传输。2)基于WMSN的多媒体图像特征融合在基于WMSN的底层网络中,采用图像、视频或者音频等多媒体传感器作为林火监控的感应设备,本文采用摄像头作为多媒体传感器,得到数据为视频图像。针对WMSN中,图像传输能量消耗大的问题,提出一种基于哈希编码的图像特征融合算法,进行火焰图像的识别,以此为依据,丢弃无效的、非火焰的图像,以减少传输量,节约能源。该算法首先需要利用测试集数据进行线下学习,建立一个图像哈希码库作为图像识别检索的依据;其次,提取待识别图像的特征并对其进行哈希编码,得到图像对应的哈希码;最后计算图像哈希码之间的汉明距离,进行检索和识别,得出是否有火灾发生。实验结果表明:基于数据传输概率阈值的数据融合算法,可以减少大约34%的传输能量消耗;基于哈希编码的图像特征融合算法,图像火焰识别的准确率可达约94.12%,高于SVM、BP神经网络和稀疏表示的火焰识别算法。基于本文提出的信息融合算法,在保证林火监控及时准确的基础上,有效减少了底层网络中无效数据的传输,从而减少了网络中传输能量的消耗,增加了网络带宽的利用率,延长了网络的生命周期。更多还原

【Abstract】 Forest resources are important natural resources, the occurrence of forest fire will cause serious resources and economic losses for the state, so, using advanced technology for intelligent monitoring of forest fires, is very important to protect the forest resources against the loss. In recent years, as constantly-developing the Internet of Things technology, it also makes application filed to expand, and one of the application is intelligent monitoring of forest fire.Wireless Sensor Network(WSN) and Wireless Multimedia Sensor Network(WMSN) as the underlying of the Internet of things, they own some characteristics, for example wide coverage, low cost, high accuracy of data, and can effectively apply in the intelligent monitoring of forest fire. But, the WSN and WMSN belong to resource-constrained network, the power the power supply equipment, computing power and bandwidth of them are limited. However, in the fire monitoring system based on the Internet of Things, there are a large number of redundant data in the underlying network, sensors can produce a large number of monitoring data, and the data transmission can consume vast network and computing resources, it seriously influence the service life of the sensor network.In this paper, aiming at the problems of WSN and WMSN, a valid information fusion algorithm was proposed, and it can reduce the transmission of data in the underlying network, save energy and improve the efficiency of monitoring and early-warning.1) The data fusion algorithm of scalar sensor based on WSNAccording to the characteristics of forest fire, we proposed a data fusion algorithm based on the transmission probability threshold. In the underlying network of WSN, it uses temperature sensor, humidity sensor, infrared sensor and smoke sensor as forest fire monitoring sensors. First, this algorithm uses the weighted average algorithm to calculate the weight coefficient of the sensor’s measurement. Second, the algorithm uses logistic regression model to get fire probability of this node. Third, compare the fire probability of this node and the threshold, if the fire probability is greater than the threshold, sent to the cluster node. Otherwise, not sent. Thus, it effectively reduces the transmission of the invalid data in the underlying network.2) The feature fusion of multimedia image based on WMSNIn the underlying network based on WMSN, it uses multimedia sensors such as image, video or audio induction as forest fire monitoring equipment. In this paper, we use the video as the image data. Aiming at the problem of image transmission energy consumption, we proposed image feature fusion algorithm based hash code, for flame image recognition. On this basis, the invalid and not flame images are discarded, in order to reduce data transmission and save energy. First, this algorithm needs to offline learning, using the test dataset, establishes the database of image hash code as the basis of image retrieval. Second, the features of the image are extracted and hash-coded. Third, we calculate the hamming distance between images’ hash code, then based on the hamming distance, the image is retrieved and identified.The experimental results show that: based on the data transmission probability threshold of the data fusion algorithm, it can reduce about 34% of the transmission energy consumption. The image feature fusion algorithm based on hash coding, the accuracy of flame recognition of this algorithm is 94.12%, and it is higher than other flame recognition algorithms, based on the SVM, BP-neural network and sparse representation. In this paper, based on the information fusion algorithm, it can effectively reduce the transmission of the invalid data in the underlying network, reduce the energy consumption of the network transmission, increase the utilization of the network, prolong the network lifetime, on the basis of guaranteeing the accuracy and timeliness of the fire monitoring.更多还原