针对传统方法的谱图预检索后参考谱图数量仍然较多,造成检索效率低和精检索时因不能利用谱图细节特征而易出现假阳性或假阴性的问题,提出了一种基于小波变换的预检索和精检索的组合匹配算法.首先利用小波变换的轮廓部分进行预检索,然后利用小波变换的细节部分进行精检索;确定了预检索时小波分解的层数为1,相似度的阈值为0.6~0.7,而精检索时小波分解层数为3;通过在相似度计算中增加小波变换的细节成分,以提高检索精度;同时根据小波变换分解尺度与质谱信号峰宽的关系,依据质谱仪的分辨率自适应地确定离散小波变换的分解层数.采用该方法在NIST08标准参考谱库(191 000张谱图)中检索了1 000多张谱图的准确性.实验表明:与加权点积方法相比,匹配的准确度平均提高了24.9%,与Stein和Scott组合方法相比,匹配的准确度平均提高了1.9%;通过质谱仪得到样品的实测谱图在参考谱库中的检索结果表明,该算法能有效提高谱图匹配的相似度和准确度,与Stein和Scott组合方法相比,甲基磷酸二甲酯的谱图相似度提高了7.6%,全氟三丁胺的谱图相似度提高了20.7%,不同浓度八氟奈的谱图相似度平均提高了14.9%,尤其在分析低浓度样品时取得了很好的效果;采用该方法处理同分异构体化合物,如邻二甲苯数据时,与Stein和Scott组合方法相比提高了相似化合物的选择性.
Aim to the problems of the number of reference spectrum was still more after presearch and false positive or false negative because of traditional method couldn't use the detail information effectively in the identification process,a composed matching method of spectrum pre-search and identification based on wavelet transform was presented.It used the approximation of the wavelet in spectrum pre-search and the detail in spectrum identifi cation.It presented that the decomposition level was 1 and the similarity threshold was 0.6-0.7 in the second pre-search,while the decomposition level was 3 in the identification.The detail information of wavelet was added in the similarity calculation to improve accuracy.According to the resolution of spectrometry,the decomposition level was setted based on the relation between peak width and scale of wavelet.In the accuracy experiment,1 000 spectra was searched in the NIST 08 reference library(191 000 spectra) using the presented method,compared with weighted dot-product and Stein and Scott composed methods,the accuracy was improved by 24.9% and 1.9% in average.Searching the spectrum of samples through the spectrometry in the reference library,the results showed that it could improve the similarity and accuracy,and compared with that of Stein and Scott composed method,the similarity value of DMMP was improved by 7.6%,and that of perfluorotributylamine spectrum was 20.7%,and that of different concentration's octafluoronaphthalene spectrum was improved by 14.9% in average,and especially acquired good result in the low concentration samples.Using the presented method to process the O-xylene spectra could improve the selectivity of similar compound.