软件人工嗅觉图像技术研究  

Software based artificial olfactory map technique

作　　者：陈铠[1,2] 秦国军[1] 胡茑庆[1] 李磊[1] 

机构地区：[1]国防科技大学机电工程与自动化学院,长沙410073 [2]中国卫星海上测控部,江阴214431

出　　处：《仪器仪表学报》2008年第12期2504-2508,共5页Chinese Journal of Scientific Instrument

基　　金：863高科技项目(2006AA05Z141);国家部委级资助项目

摘　　要：模式识别是人工嗅觉的重要组成部分和关键技术之一,虽然目前该领域对具体的识别算法进行了较多的研究,但对数据预处理及实验数据样本总体信息完整性描述着墨较少。为此,本文模拟光寻址电位传感器构建人工味觉图像的原理,提出了软件人工嗅觉图像技术。该方法从"传感器-时间-样本"三维结构样本出发,在对实验数据样本总体进行重组的基础上,使用独立分量分析技术削减冗余信息,生成基于软件的嗅觉图像,并产生嗅觉记忆。实验结果表明,该方法较好地保留了样本的总体信息,在嗅觉图像基础上,可以采用简单的分类器对气味进行正确分类,算法复杂度低。Pattern recognition is a key technology in artificial olfaction. Many algorithms have been researched and investigated in this field; however the details about information integrity and data preprocessing technique are described rarely. A software-based artificial olfactory map technique mimicking artificial gustatoW maps based on light addressable potential sensor is proposed in this paper. Starting from reconstructing the ＂sensor-time-sample＂ 3-dimensioned experimental data and removing the redundant information with independent component analysis （ICA）, the olfactory maps, as well as olfactory memory based on software are generated. Experiment results indicate that the technique can retain the entire information of the sample properly, has low calculation complexity ; and based on artificial olfactory maps, a simple classifier can be used to recognize the odor easily.

关 键 词：人工嗅觉 传感器阵列 特征提取 独立分量分析 

分 类 号：TM938.83[电气工程—电力电子与电力传动]