论文部分内容阅读
高光谱数据以其高光谱分辨率和多而连续的光谱波段为预测土壤重金属污染提供了有力工具,但波段选择方法与光谱分辨率的影响不容忽视。利用实验室测定的181个土壤光谱样本数据,利用逐步回归法进行土壤Cu含量反演的波段选择,进而利用偏最小二乘方回归PLSR方法建模,分析了波段数对Cu含量反演的影响;此外,采用高斯响应函数重采样方法,探讨了光谱分辨率降低对反演精度的影响。实验表明,预测重金属元素Cu含量的最佳波段数为10个,模型可决系数R2=0.7523,拟合均方根误差RMSE=0.4699;预测Cu含量的最佳光谱采样间隔为32 nm,R2=0.7028,RMSE=0.5147。该结果可能为将来设计低廉实用的高光谱卫星传感器提供指标论证,为模拟卫星传感器波段预测土壤重金属含量提供理论依据。
Hyperspectral data provide a powerful tool for predicting heavy metal pollution in soil with its high spectral resolution and multiple and continuous spectral bands. However, the influence of band selection method and spectral resolution can not be ignored. Based on the data of 181 soil spectral samples measured in the laboratory, stepwise regression was used to select the band of soil Cu content inversion, and then the PLSR method was used to model the influence of the band number on the Cu content inversion In addition, Gaussian response function resampling method was used to investigate the effect of spectral resolution reduction on inversion accuracy. The experimental results show that the optimal wavelength band for predicting the content of Cu in heavy metal is 10, the model coefficient of determination is R2 = 0.7523, and the root mean square error of fitting (RMSE) is 0.4699. The optimal spectral sampling interval for predicting Cu content is 32 nm, R2 = 0.7028, RMSE = 0.5147. The results may provide indicators for the design of low-cost hyperspectral satellite sensors in the future to provide a theoretical basis for the prediction of heavy metal content in soil by simulating satellite sensor bands.