基于高光谱的水体BOD含量模拟估算

Simulation Estimation of BOD Content in Water Based on Hyperspectra

高光谱技术由于满足连续性与光谱可分性的要求,具有能够区别同一种地物不同类别的能力,且光谱数据获取速度快,操作简易,在监测水体分布状况、水体指标上具有突出成就。生化需氧量BOD是评价水污染的重要指标,现行常规的测量方法为五日培养法,这种方法消耗试剂、操作复杂、受干扰因素多、测定时间长、不能及时反映水质变化、无法及时有效地预警突发水污染事件,鉴于传统方法的缺点,探索基于高光谱技术的水体BOD含量的估算和反演对水质评定具有重要意义。以西安地区三处地表水为研究区,共计60处试验点,每处试验点重复测定10次光谱与BOD值,取平均值作为原始光谱,利用Person相关系数法筛选光谱与BOD值的敏感波段,并使用主成分分析与最小二乘法消除光谱指标的多重共线性,建立水质BOD指标的多元线性回归模型与偏最小二乘回归模型。研究结果如下:(1)BOD敏感波段大体分布于600~900 nm,共筛选出了35个显著相关的原始光谱指标,其中758 nm相关系数绝对值最高(0.418);(2)经由主成分分析降维得出的Z_(1)和Z_(2)与BOD指标的多元线性回归模型精度较好(R^(2)=0.565,RMS E=0.007),且主成分分析中可以明显区分0~0.2与0.4~0.6 mol·L^(-1) BOD浓度;(3)光谱指标与BOD指标构建偏最小二乘回归模型的精度R^(2)高达0.896,RMSEP=0.7469(留一交叉法均方根误差);jack.test检验发现628 nm对反演水体BOD含量的影响极其显著,889与893 nm波段对其影响较为显著;(4)根据模型拟合精度,筛选的最优的BOD反演模型为偏最小二乘回归模型,对偏最小二乘模型进行精度检验,精度较好(R^(2)=0.81)。基于以上试验结果,提出了一种基于偏最小二乘法高光谱水质BOD参数的反演方法,为水质BOD参数动态检测提供了新方法。

Due to the requirement of continuity and spectral separability,hyperspectral technology has the ability to distinguish different types of the same ground object,and the spectral data acquisition speed is fast,and the operation is simple.Spectral analysis has made outstanding achievements in monitoring water distribution and water indicators.Biochemical oxygen demandis one of the important indicators to evaluate water pollution,the current conventional measuring method for 5 culture method,and this method consumes reagent,complicated operation,more interference factors,determination of time is long,can not reflect the water quality changes in time,can’t early warning of emergent water pollution events in a timely and effective manner,in view of the traditional methods of faults,explore the content of water,BOD estimation based on the technology of hyperspectral and inversion for water quality assessment is of great significance.This test three surface water in xi’an area as the research area,a total of 60 sites,each site repeat 10 times spectra and the BOD value,average as an original spectrum and the BOD value,Person correlation coefficient method is used to filter the spectrum and the BOD value of sensitive wavebands,and principal component analysis and least square method are used to eliminate spectral index of multicollinearity,BOD water quality index of the multivariate linear regression model and partial least squares regression model.The results were as follows:(1)the BOD sensitive bands were generally distributed at 600~900 nm,and a total of 35 original spectral indicators with significant correlation were screened out,of which the absolute value of the correlation coefficient of 758 nm was the highest(0.418).(2)the accuracy of multiple linear regression model of Z_(1),Z_(2) and BOD indexes obtained by principal component analysis(R^(2)=0.565,RMSE=0.007)is good,and the BOD concentration of 0~0.2 and 0.4~0.6 mol·L^(-1) can be clearly distinguished in the principal component analysis.(3)partial least-squares regression between spectral index and BOD index shows that the model accuracy R^(2) of the partial least-squares regression model is up to 0.896,RMSEP=0.7469(root mean square error with one crossing method).By jack test,it is found that 628 nm has a very significant influence on the BOD content of inversion water body,and the bands of 889 and 893 nm have a significant influence on it.(4)according to the model fitting accuracy,the selected optimal BOD inversion model is the partial least squares regression model,and the accuracy of the partial least squares model is verified to be good(R^(2)=0.81).Based on the above test results,an inversion method based on partial least squares hyperspectral BOD parameters of water quality is proposed,which provides a new method for dynamic detection of water quality BOD parameters.