基于特征区域下凸点提取的藻类荧光光谱波长选择方法

藻华现象的频繁发生严重影响了海洋环境和人类的生产活动，因此对水体浮游植物的监测十分重要。三维荧光光谱被广泛应用于水体浮游植物中藻类的群落组成分析和浓度定量分析，然而三维荧光光谱数据中的信息冗余给藻类定性定量分析带来了一定的影响。针对光谱信息冗余问题，提出了特征区域积分与凸点提取相结合的三维荧光光谱波长选择方法。以抑食金球藻、细长聚球藻、小球藻为研究对象，采用Savitzky-Golay卷积平滑法对三维荧光光谱进行预处理，解决了因外界因素造成的光谱噪声问题，采用马氏距离法剔除三维荧光光谱数据集中的异常光谱样本，运用浓度残差法剔除三维荧光光谱数据集中的异常浓度值样本，然后通过偏最小二乘回归模型的内部交叉验证均方根误差衡量不同特征区域下凸点的可靠性进行波长变量的选择。为验证波长筛选方法的有效性，对三种藻类建立偏最小二乘回归模型，以内部交叉验证决定系数（R2）、内部交叉验证均方根误差（RMSECV）作为模型评价指标。与全光谱数据建立的回归模型进行了比较，抑食金球藻、小球藻、细长聚球藻的波长变量由全谱的1071个分别减少到77个、75个、67个，R2分别提高了0.016 4，0.002和0.032 4，RMSECV分别降低了1.8×105，2.0×105，2.6×105。与UVE方法相比，抑食金球藻、小球藻、细长聚球藻的波长变量分别减少了599个、357个、317个，R2分别提高了0.014 5，0.000 4和0.012 3，RMSECV分别降低了1.6×105，7.0×104和1.6×105。经过该方法进行波长变量选择后，减少了冗余信息，提高了模型预测能力。

Wavelength Selection Method of Algal Fluorescence Spectrum Based on Convex Point Extraction From Feature Region

The frequent occurrence of algal bloom seriously affects the Marine environment and human production activities, so it is very important to monitor the phytoplankton in water.3D fluorescence spectroscopy has been widely used in the analysis of algae community composition and the quantitative analysis of algae concentration in water phytoplankton. However, the information redundancy in 3D fluorescence spectrum data has significantly impacted the qualitative and quantitative analysis of algae.In order to solve the problem of spectral information redundancy, a new wavelength selection method of 3D fluorescence spectrum based on the combination of feature region and convex point extraction is proposed.Taking Aureococcus anophagefferens, Chlorella Vulgaris, and Synechococcus elongatus as the research object, the Savitzky-Golay convolution smoothing method was used to preprocess the 3D fluorescence spectrum to solve the problem of spectral noise caused by external factors. The Mahalanobis distance method was used to eliminate the abnormal spectral samples in the 3D fluorescence spectrum data set.The residual concentration method was used to eliminate the abnormal concentration value samples in the 3D fluorescence spectrum data set.Then the reliability of the convex points under different characteristic regions was measured by the root mean square error of cross-validation (RMSECV) of the PLS regression model, and the wavelength variable was selected. In order to verify the effectiveness of the wavelength selection method, the PLS regression model was established for the three algae species, and the determination coefficient (R2) and root mean square error of cross-validation (RMSECV) were used as the evaluation indexes of the model. Compared with the regression model established with the full spectrum data, the wavelength variables of Aureococcus anophagefferens, Chlorella Vulgaris, and Synechococcus elongatus respectively decreased from 1 071 to 77, 75 and 67, and R2 respectively increased by 0.016 4, 0.002 and 0.032 4. RMSECV was respectively reduced by 1.8×105, 2.0×105 and 2.6×105. Compared with the UVE method, the wavelength variables of Aureococcus anophagefferens, Chlorella Vulgaris, and Synechococcus elongatus were respectively reduced by 599, 357 and 317, and R2 was respectively increased by 0.014 5, 0.000 4 and 0.012 3, RMSECV was respectively decreased by 1.6×105, 7.0×104 and 1.6×105. After the selection of wavelength variables by the method of feature region combined with convex point extraction, the redundant information is reduced, and the model’s prediction ability is improved.