基于coif2小波的浮游植物活体三维荧光光谱识别技术研究

小波分析技术是提取不同门类以及种属水平上浮游植物的三维荧光光谱特征的有效手段,利用coif2小波函数对分属于7个门,30个属的37种我国近海常见的浮游植物的三维荧光光谱进行小波分解,小波分量和尺度分量作为浮游植物备选荧光特征谱,通过Bayes分析确定第3层尺度分量作为浮游植物门类特征光谱,第3层尺度分量和第2和第3层小波分量的组合作为浮游植物属特征谱。对获得的浮游植物荧光特征谱进行系统聚类分析,得到37种浮游植物门类水平上的107条和属水平上的155条浮游植物荧光标准谱,组成浮游植物荧光标准谱库。在此标准谱库的基础上,利用非负最小二乘法解析的多元线性回归建立浮游植物三维荧光光谱识别技术。该技术对1776个单种藻样品和384个混合藻样品进行识别分析,单种浮游植物样品在门类水平上的识别正确率为98.1%,属水平上的识别正确率为97.0%;浮游植物混合样品中的优势种在门水平上的识别正确率分别为94.8%,在属水平上的识别正确率为92.7%。

Research on the 3D Fluorescence Spectra Differentiation of Phytoplankton by Coiflet2 Wavelet

In the present paper,the authors utilize the wavelet base function eoiflet2 (coil2) to analyze the 3D fluorescence spectra of 37 phytoplankton species belonging to 30 genera of 7 divisions,and these phytoplankton species include common species frequently causing harmful algal blooms and most predominant algal species in the inshore area of China Sea.After the Rayleigh and Raman scattering peaks were removed by the Delaunay triangulation interpolation,the fluorescence spectra of those phytoplankton species were transformed with the coiflet2 wavelet,and the scale vectors and the wavelet vectors were candidate for the feature spectra.Based on the testing results by Bayesian analysis,the 3rd scale vectors were the best feature segments at the division level and picked out as the fluorescence division feature spectra of those phytoplankton species,and the group of the 3rd scale vectors,the 2nd and 3rd wavelet vectors were the best feature segments at the genus level and chosen as the fluorescent genus feature spectra of those phytoplankton species.The reference spectra of those phytoplankton species at the division level and that at the genus level were obtained from these feature spectra by cluster analysis,respectively.The reference spectra base for 37 phytoplankton species was composed of 107 reference spectra at the division level and 155 ones at the genus level Based on this reference spectra base,a fluorometric discriminating method for phytoplankton populations was established by multiple linear regression resolved by the nonnegative least squares.For 1 776 samples of single phytoplankton species,a correct discriminating rate of 97.0% at genus level and 98.1% at division level can be obtained;The correct discriminating rates are more than 92.7% at the genus level and more than 94.8% at the division level for 384 mixed samples from two phytoplankton species.