小波变换的EDXRF光谱金属组分特征峰位置识别

摘要

In this paper ,the accurate identification problem of energy dispersive X-ray fluorescence (EDXRF) characteristic peak position was studied.Based on the characteristic spectra character of the different metal components ,the choosing rule of the characteristic spectra was analyzed.According to the theories of singular value analysis and modulus maxima ,the extraction method of modulus maxima was analyzed which based on the wavelet decomposition coefficients of characteristic spectra.Moreo-ver ,the feature of the characteristic spectra wavelet decomposition coefficients and their propagation were analyzed in detail.The method of the interval characteristic peak selection was put forward based on the propagation of modulus maxima.And this method was applied to the actual measurement spectra.The result showed that the wavelet transform of four levels was applied to full energy spectra data using the basis function of bior4.4 wavelet.For the full energy spectra ,the phase step influence of the some superimposed noise could be eliminated using the propagation of modulus maxima.In order to increase the identification probability of characteristic spectra ,the decomposition coefficients were compressed which were less than the threshold value.In addition ,667 peak positions were identified for the fourth level wavelet decomposition coefficients of EDXRF spectra which were not processed.186 peak positions were identified when they were compressed.Then the method of interval characteristic peak selection using modulus maxima propagation feature was applied and the initial value of the screening interval was set 600 eV. The identified result of the characteristic peak position was 27.The experimental result showed that the accurate rate of peak lo-cation identification was enhanced effectively.%主要研究X射线荧光光谱金属组分特征谱位置的确定.依据不同金属组分的特征谱特性,分析了特征谱的选取规律,在奇异值分析理论和模极大值理论的基础上,分析了基于特征谱小波分解系数的模极大值提取方法,在不同分解尺度下的特点及其传播特性,提出了基于模极大值传播的区间特征峰筛选方法,并对实际测量光谱进行了实验分析.结果表明:利用bio r4.4小波作为基函数对实验测量的全能谱数据进行4层小波变换,利用模极大值传播特性,可以消除全能谱上叠加的部分噪声对光谱分析造成的阶跃影响;为提高特征峰的位置识别概率,对小波变换中小于给定阈值的分解系数进行压缩,将实验获取的X射线荧光全能谱第4层小波分解系数直接进行特征峰识别,得到的677个峰值位置,压缩到186个;在此基础上,再采用模极大值传播的区间特征峰筛选方法,筛选区间初始值设置为600 eV,经识别得到的特征峰峰值位置仅为27个,识别准确率得到有效提高.