检测室内有害气体得到的红外光谱为混合有害气体的红外光谱,针对吸收谱带相互交叠的混合气体定性定量不容易的问题,提出基于特征矩阵联合近似对角化(joint approximative diagonalization of eigenmatrix,JADE)的特征提取方法,该方法通过分析数据的高阶统计量信息,充分挖掘原始数据隐含的信息,以便准确地区分出混合气体中各物质的光谱,同时应用基于正则理论的支持向量机(SVM)对提取出来的独立信号源建立多维数据定量分析的模型.实验结果表明,混合气体中各组分的定量分析相关系数均保持在0.9991以上,验证了该特征提取方法的准确性.%The infrared spectrum obtained by indoor air pollution monitor is a variety of harmful mixture gas and absorption bands of mixture gas overlap makes qualitation a difficult question. A feature extraction method based on joint approximative diagonalization of eigenmatrix (JADE) is proposed. The method can fully mine implicit information in the original data by analyzing the Higher-order statistics information so that we can separate mixture gas spectrum into each material's spectrum. SVM (support vector machine) based on the regular theory is applied to establish a multi-dimensional data quantitative analysis model by the extracted independent source. The experimental result shows that the relevant factors of mixture gas component quantitative analysis are maintained at 0.9991, which proves the accuracy of this feature extraction method.
展开▼
