一种结合粗糙集与神经网络的多传感器信息融合算法  

AN ALGORITHM OF MULTI-SENSOR INFORMATION FUSION BASED ON THE COMBINATION OF ROUGH SET AND NEURAL NETWORK

作　　者：刘决仕[1] 金晶[1] 陈华曦[1] 

机构地区：[1]上海交通大学计算机科学与工程系,上海200240

出　　处：《计算机应用与软件》2009年第2期14-16,34,共4页Computer Applications and Software

基　　金：国家自然科学基金项目(60372078)

摘　　要：近年来,神经网络被广泛应用于多传感器信息融合。但是当传感器数量庞大时,过高的输入神经网络的信息维数会导致神经网络训练速度下降,甚至不收敛。针对上述问题,对传统的基于神经网络的融合算法进行了改进,利用粗糙集的冗余数据约简算法,剔除部分传感器的输入,同时将剩余的传感器信息重新组合,形成维数较小的数据分别训练,从而避免了输入数据维数过高带来的问题,较之于传统算法,算法在训练阶段的迭代次数等时间性能以及融合阶段的准确性两个方面均有所提高。Recently, neural networks have been widely used in muhi-sensor information fusion. However, training becomes slow, even cannot converge due to a rapid increase in input dimension when there＇ re large number of sensors. This paper proposed an improved fusion algorithm based on neural networks to address that issue. Using reduction method for redundant data in rough set theory, the proposed algorithm eliminates some sensors＇ inputs and reorganizes rest inputs to form smaller data dimensions for separate training, avoiding disadvantages brought by high dimension. Experiments indicated that compared with traditional method, the proposed algorithm achieves impactful progress both in time performance including iteration times and so on in the training phase and the accuracy in the fusion phase.

关 键 词：粗糙集 神经网络 信息融合 

分 类 号：TP274.2[自动化与计算机技术—检测技术与自动化装置] TP391.41[自动化与计算机技术—控制科学与工程]