论文部分内容阅读
采用激光诱导击穿光谱技术与因子分析以及BP神经网络技术相结合,对5类共9种标准样品进行岩性分类与相互区分.根据样品的主要元素含量选取Si,Al,Ca,Fe,Mg,K,Na共7种元素的波峰构成特征谱.每种元素均选取一个峰作为研究对象,根据峰的形状及大小确定每个峰的波长取值范围.利用因子分析对全谱和特征谱分别进行主成分分析,再将得到的全谱主成分和特征谱主成分以及全谱与特征谱分别输入BP神经网络进行样品的岩性分类与相互区分.在以上4种情况下,样品的岩性分类进行BP神经网络分析,4种结果中以特征谱的识别率为最高,是98.89%;样品的相互区分进行BP神经网络分析,也以特征谱的识别率为最高,是98.89%.实验结果表明,对全谱进行特征提取后得到的特征谱,可以代表全谱进行因子分析和BP神经网络分析,且能更准确与高效地完成样品分类与相互区分.
The combination of laser-induced breakdown spectroscopy, factor analysis and BP neural network technology were used to classify and distinguish lithology of 9 kinds of standard samples from 5 kinds of samples.According to the main elements of samples, Si, Al, Ca, Fe, Mg , K and Na.A peak is selected for each element, and the wavelength range of each peak is determined according to the shape and size of the peak.The full-spectrum and the characteristic spectrum Principal Component Analysis (PCA) and Principal Component Analysis (PCA), respectively, and the principal components, the whole spectrum and the characteristic spectrum of the full spectrum principal component and the characteristic spectrum were input into the BP neural network respectively to classify and distinguish the lithology of the samples. In the above four cases, According to the results of BP neural network analysis, the recognition rate of the characteristic spectrum among the four kinds of results was the highest (98.89%), and BP neural network analysis was used to distinguish the samples from each other, and the recognition rate of the characteristic spectrum was the highest (98.89%). The results show that the feature spectrum obtained after the feature extraction of the full spectrum can be used to perform factor analysis and BP neural network analysis on the whole spectrum, and to classify and distinguish samples more accurately and efficiently.