基于高光谱定量反演模型的污水综合水质评价

Comprehensive Evaluation of Waste Water Quality Based on Quantitative Inversion Model Hyperspectral Technology

为改善高光谱遥感对污水水质信息状况定量反演模型的预测评价效果,以陕西某污水处理厂采集的污水样品为研究对象,采用主成分分析法(Principal component analysis,PCA)对污水水质进行综合评价,获取水质评价的综合评价因子,同时利用ASD Field Spec 3型高光谱仪获取污水的原始光谱,经过数据预处理和不同数学变换后,共获取了4种光谱指标:平滑后光谱反射率(SG)、倒数之对数(LR)、标准正态化(SNV)和去包络线(CR)。分别采用偏最小二乘回归法(Partial least squares regression,PLSR)、逐步回归法(Stepwise regression,SR)、极限学习机法(Extreme learning machine,ELM)构建了基于水质综合评价因子的高光谱水质反演模型,并对反演结果进行精度验证与比较。结果表明,本组水样的平滑后光谱数据和经过标准正态化变换的光谱数据建模具有较好的建模效果,其建模的预测RPD均在2. 5以上;在3种模型中,PLSR模型和ELM模型均具备很好的建模预测效果;逐步回归法的建模效果较PLSR模型和ELM模型有所下降,但是其SG-SR、SNV-SR模型的Rc^2均在0. 8以上、Rp^2均在0. 85以上,RPD均在3. 0以上,证明其仍拥有很好的反演预测效果,且进行了特征波段的优选,实现了对模型的优化;SNVSR-ELM(Rc^2=0. 956,Rp^2=0. 954,RMSE=0. 500,RPD=4. 651)为最佳模型,SNV-SR-ELM模型的建立为高光谱反演水质模型的优化、污水水质的快速监测和综合评价提供了途径。

A comprehensive inversion of the water quality information of sewage water was realized through the combination of hyperspectral technology and water quality comprehensive evaluation method.Taking the sewage sample collected by a sewage treatment plant in Shaanxi as the research object,principal component analysis( PCA) was used to comprehensively evaluate the sewage water quality to obtain a comprehensive evaluation factor for water quality. At the same time,the original wastewater spectrum was obtained by the ASD FieldSpec 3 hyperspectral instrument. After data preprocessing and different mathematical transformations,four spectral indices were obtained: spectral reflectance( SG),reciprocal logarithm( LR),standard normal variable( SNV) and continuum removed( CR). Based on partial least squares regression( PLSR),stepwise regression( SR) and extreme learning machine( ELM), a hyperspectral model of inversion water quality comprehensive evaluation factor was constructed. The results showed that the original spectral data of this group of water samples and the spectral data modeling by standard normalization transformation had good modeling results,and the prediction effect RPD of the model was above 2. 5. Among the three models,the PLSR model and the ELM model had good modeling prediction effects,while stepwise regression modeling results were declined compared with PLSR model and ELM model,the Rc^2 and Rp^2 of the REF-SR and SNV-SR models were all above 0. 8 and 0. 85,and the RPD was above 3. 0,which still had a very good inversion prediction effect,and it achieved the optimization of the model and the optimization of the characteristic band,and SNV-SR-ELM( Rc^2= 0. 956,Rp^2= 0. 954,RMSE = 0. 500,RPD = 4. 651) was the best model. The establishment of SNV-SR-ELM model provided a way for the optimization of hyperspectral inversion water quality model and the rapid evaluation of sewage water quality.