【作者】 张宇林；

【导师】 徐保国；

【作者基本信息】 江南大学 ， 轻工信息技术与工程， 2009， 博士

【摘要】 多传感器数据融合技术首先是从军事领域发展起来的,并且己经发展成为一个新的学科方向和研究领域。本文介绍了数据融合系统的基本理论、发展及研究现状,研究了数据融合中的不确定方法,结合在研的国家“863”重大项目——车载农田土壤信息快速采集关键技术与产品研发,构建土壤参数融合模型,在智能计算理论的基础上,对多传感器数据融合算法提出了新的思路,并进行了深入的研究。对基于神经网络与信任度函数的多传感器数据融合算法进行了研究,构造信任度函数矩阵,采用该矩阵度量各传感器测得数据之间的综合信任程度,合理分配测得数据在融合过程中的权重,作为BP神经网络的输入,经过训练好的网络,取得良好的融合效果,以削弱白噪声对测量精度的影响。提出了两种基于粒子群优化算法(PSO)的传感器融合方法。一种方法是对粒子群优化算法中的固定惯性权重进行改进,分析了惯性权值因子在粒子群优化算法中的作用,在现有线性递减权值方法上,提出一种非线性权值递减策略,并将其尝试性的运用到多传感器数据融合的领域,运用该算法对加权因子进行调整,得到良好的融合效果。另一种是基于量子空间的粒子群算法(QDPSO)和BP神经网络的多传感器融合算法,经过QDPSO训练的BP神经网络具有较好的稳定性和收敛性,将其运用于多传感器数据融合中,通过仿真取得了比常规算法更高的精度,是一种较有潜力的融合算法。研究小波变换的特性,提出了基于小波包变换的多分辨率多传感器信息融合模型。利用小波变换的理论,研究小波在像素级、特征级融合中的新算法。根据已经建立好的土壤信息融合模型,结合测量得到的各种农田参数,将研究的融合算法运用于具体的数据处理中,对土壤含水率、电导率等参数进行实际分析,从海量传感器信息中,提取有价值的农作物生物物理和生物化学参数,以指导实地农业生产和管理,从而提高农作物的产量和质量,为进一步实现精准农业提供理论依据和实践参考。更多还原

【Abstract】 Multi-sensor data fusion technology was developed firstly from the military area, and had been developed into a new subject direction and research field.In this paper, the definition and the model of data fusion were introduced together with its development and research status.Studied the uncertain methods of the data fusion.Combined with the research at the National“863”major projects—the key information and product development of express collection of soil information, Constructed the model of soil parameters.On the base of the theory of computational intelligence, New ideas were put forward based on multi-sensor data fusion algorithm and the thorough research had been done.Do some research on data fusion model and algorithm combined with BP and belief function.Build belief function matrix and use it to measure the belief degree of sensor data.Distribute the weight of the fusion process reasonably which is the input of BP neural network.The fusion effect is good through the best trained network which weaks the white noise of measurement accuracy while at the same time a number of sensor observation value were compressed into an optimal fusioned data.Two sensor fusion algorithms were given based on particle swarm optimization (PSO).One method was to improve the inertia weight of PSO where the role of the inertia weight factor in PSO was analysed. A nonlinear strategy for DIW was put forward to the field of the multi-sensor fusion which can estimate the weight factor of the data fusion algorithm and get closer to the ture value. The other method is to use multi-sensor data fusion algorithm based on QDPSO-BP network, throgh which trained BP neural network and got good stability and convergence.Higher accuracy would be gotten by using it in the simulation.It is a potential data fusion method for multi-sensors.Research the character of wavelet transform.The multi-resolution and multi-sensor data fusion model was proposed based on wavelet packet.Research the new multi-sensor data fusion algorithm in the pixel level, feature level by using the wavelet theory.Apply the data fusion algorithm to specific data processing combined the measurement parameters of the various fields. Analysis the soil moisture content, conductivity and other parameters actually. Extract the biochemistry and biophysics parameters from valuable crops and guide the agricultural production and management so as to enhance crop yield and quality.Provide a theoretical base on precision agriculture for the further implementation.更多还原