一种新的证据组合规则应用于序列图像目标识别  

New Combination Rule in Evidence Theory Applied to Target Recognition of Sequential Images

作　　者：杨阳[1] 程咏梅[1] 潘泉[1] 张洪才[1] 苗壮[1] 

机构地区：[1]西北工业大学自动化学院,陕西西安710072

出　　处：《计算机应用研究》2007年第4期170-172,共3页Application Research of Computers

基　　金：国家自然科学基金资助项目(60372085;60404011)

摘　　要：提出了一种新的证据理论组合规则,并结合BP神经网络和证据推理对序列图像目标识别问题进行应用。以修正的Hu不变矩为图像特征,利用数据融合的思想对来自目标的序列图像进行时间域融合处理。由BP神经网络对目标的初步识别结果构造基本置信指派函数,用该组合规则进行决策级数据融合,完成了三维飞机图像目标的识别仿真。仿真结果表明,新的组合规则结合BP神经网络的融合方法提高了三维飞机目标识别的准确性。It presented a new combination rule of Dempster-Shafer evidence theory. The question of sequential image recognition was solved based on BP Neural Network （BPNN） and Dempster-Shafer evidence theory. Firstly, modified Hu invariant moments was used as the feature of the image, and BPNN was applied to identify the target. Secondly, basic belief assignment function was constructed through the output of the BPNN, and afterwards the new combination rule was used to finish the deci- sion data fusion. Lastly, recognition of 3D airplane images was completed. The simulation results showed that this method is effective.

关 键 词：目标识别 D-S证据理论 序列图像 反向传播神经网络 组合规则 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]