基于约束独立成分分析的多组分污染气体混叠峰识别

Identification of Multi-component Pollution Gas Aliasing Peak Based on Constrained Independent Component Analysis

大气污染监测中测得的红外光谱是混合气体的红外光谱。针对主次吸收峰严重混叠的红外混合气体特征提取和识别问题,提出了一种基于约束独立成分分析的光谱特征提取方法使谱线有效地区分开来。该方法引用峭度作为约束条件,运算时约束向量可使所对应的目标信源具有非高斯性极大的效用,提取出目标信源。并按统计度量的大小来进行排序,减少传统独立成分分析算法分离信号顺序、符号、幅度的不确定性,使得分离出来的独立分量更接近实际信源。应用支持向量机建立定量分析模型检验该算法分离效果。实验结果表明,该方法能够按需要顺序分离出混合气体光谱中的纯物质光谱。

The infrared spectrum measured in air pollution monitoring is mixed gas infrared spectrum. A method based on constrained independent component analysis is proposed to solve the feature extraction and identification where the primary and secondary absorbed peaks are seriously overlapped. The method cited kurtosis as constraints. The constraints vector lead to the correspond target have a non-Gaussian source of great utility during operation. Extracting the target signal source and according to the size of statistical measure to sort. It can reduce the uncertainty of signal sequence,symbol,amplitude separated by traditional independent component analysis. So the separated independent component is closer to the actual source. Then apply support vector machine algorithm quantitative analysis model to examine the separation effect. The experimental result indicates that this method can separate the spectrum of pure substances from mixed gas spectrum in desired order.