耦合小波变换和偏最小二乘的悬浮物浓度和浊度高光谱建模方法

针对高光谱曲线中可能存在噪声以及传统半经验方法不能有效利用全部光谱信息的问题,提出了耦合Haar小波变换和偏最小二乘的水质遥感高光谱建模方法(Haar WT-PLS)。利用该方法,对在南四湖获取的实测高光谱数据经分解尺度为3的Haar小波变换后,将原始光谱数据压缩到47个特征变量;随后利用小波变换重构的光谱数据建立了悬浮物浓度和浊度的Haar WT-PLS反演模型,并进行了验证。结果表明:Haar WT-PLS反演悬浮物浓度和浊度精度较高,验证样本的均方根误差分别为25.05 mg/L和20.10NTU,平均相对误差分别为20.36%和13.88%。通过和单波段模型、一阶微分模型和波段比值模型进行精度对比分析,本文建立的Haar WT-PLS模型反演悬浮物浓度和浊度具有较高的精度和更好的稳定性。

Hyperspectral modeling method of suspended solid concentration and turbidity based on the coupling of Haar wavelet transform and partial least squares

Aiming at the noise in hyperspectral curve and the problems that traditional semi-empirical meth-ods cannot effectively use all effective spectral information,hyperspectral modeling method of suspended sol-id concentration and turbidity based on the coupling of Haar wavelet transform and partial least squares (Haar wavelet transform) is proposed. Firstly, the original spectral data of Nansi Lake on July 22 to 23, 2014 are compressed to 47 feature variables using wavelet transform with the wavelet function Haar and de-compose scale 3. Secondly, according to the reconstruction data of spectra using wavelet transform, Haar WT-PLS inversion models of suspended solid concentration and turbidity are established and verified. The results show that inversion of suspended solid concentration and turbidity using Haar WT-PLS has higher accuracy and the root mean square errors of validation samples are 25.05 mg/L and 20.10 NTU, respective-ly. The average relative errors are 20.36 % and 13.88 %,respectively. Through the analysis and comparison of single band model, the first derivative model, band ratio models and Haar WT-PLS model, Haar WT-PLS model proposed in the paper has higher accuracy and stability to retrieve suspended solid concen-tration and turbidity.