荧光光谱成像技术结合聚类分析及主成分分析的藻类鉴别研究

为探讨快速、实时藻类检测方法，实验通过荧光光谱成像技术结合模式识别方法对不同藻类进行鉴别研究。发现藻类样本存在着显著的荧光特性，通过采集40个藻类样品的荧光光谱图像，对图像进行去噪、二值化处理，确定有效像素后，根据光谱立方体绘制每个样本的光谱曲线，将所得400～720 nm区段范围内的光谱数据作鉴别分析，再利用系统聚类分析及主成分分析两种不同的模式识别法对光谱数据进行处理。系统聚类分析结果表明： 采用欧氏距离法及平均加权法计算样本间的聚类距离，在距离L=2.452以上水平处可将样本正确分类，准确率为100%;主成分分析结果表明： 通过对原始光谱数据进行一阶微分、二阶微分、多元散射校正、变量标准化等预处理后，再对数据进行主成分分析，其中二阶微分预处理后鉴别效果最佳，八种藻类样品在主成分特征空间中独立分布。因此，利用荧光光谱成像技术结合聚类分析法及主成分分析法对藻类进行鉴别是可行的，操作简便、快速、无损。

Algae Identification Research Based on Fluorescence Spectral Imaging Technology Combined with Cluster Analysis and Principal Component Analysis

In order to explore rapid real-time algae detection methods, in the present study experiments were carried out to use fluorescence spectral imaging technology combined with a pattern recognition method for identification research of different types of algae. The fluorescence effect of algae samples is obvious during the detection. The fluorescence spectral imaging system was adopted to collect spectral images of 40 algal samples. Through image denoising, binarization processing and making sure the effective pixels, the spectral curves of each sample were drawn according to the spectral cube .The spectra in the 400～720 nm wavelength range were obtained. Then, two pattern recognition methods, i.e. hierarchical cluster analysis and principal component analysis, were used to process the spectral data. The hierarchical cluster analysis results showed that the Euclidean distance method and average weighted method were used to calculate the cluster distance between samples, and the samples could be correctly classified at a level of the distance L=2.452 or above, with an accuracy of 100%. The principal component analysis results showed that first-order derivative, second-order derivative, multiplicative scatter correction, standard normal variate and other pretreatments were carried out on raw spectral data, then principal component analysis was conducted, among which the identification effect after the second-order derivative pretreatment was shown to be the most effective, and eight types of algae samples were independently distributed in the principal component eigenspace. It was thus shown that it was feasible to use fluorescence spectral imaging technology combined with cluster analysis and principal component analysis for algae identification. The method had the characteristics of being easy to operate, fast and nondestructive.