基于分数阶微分的荒漠土壤铬含量高光谱检测

为解决高光谱检测土壤中痕量级重金属含量存在的困难,提高土壤重金属铬含量检测的准确度,利用新疆准东煤田周边168个荒漠土壤样本的重金属铬含量及其对应的高光谱数据,运用分数阶微分算法进行光谱数据预处理,最后利用全部波段进行偏最小二乘建模并进行可视化分析,旨在探讨分数阶微分预处理在高光谱数据估算荒漠土壤重金属铬含量的可能性。结果表明:原始光谱与吸光率变换的分数阶微分模型均在1.8阶微分处达到了最好的精度效果。吸光率变换1.8阶微分模型为最优模型,模型的校正均方根误差为7.68 mg/kg,R_c~2=0.83,预测均方根误差为8.39 mg/kg,R_p~2=0.78,相对分析误差为2.14。最后利用铬含量实测值与光谱预测值通过反距离加权法插值获得研究区土壤重金属铬含量的空间分布,说明利用该方法对土壤重金属铬含量定量检测并进行大尺度的空间分布反演在一定程度上是可行的,为荒漠土壤重金属污染状况的高光谱检测提供了一定的科学依据和技术支持。

Spectral Detection of Chromium Content in Desert Soil Based on Fractional Differential

To solve the problem in prediction of soil heavy metal content at trace levels by hyperspectral data and improve the accuracy of prediction in soil chromium (Cr) content, fractional order differential algorithm was brought in to preprocess hyperspectral data. With 168 samples of soil taken from the open coalmine area in Eastern Junggar Basin, China, the soil heavy metal Cr contents and the reflectance of these samples were measured by indoors experiments. The hyperspectral data were preprocessed by using fractional order differential algorithm, all of the wavelengths among 401~2400nm were used to calibrate the hyperspectral estimation models of soil Cr content by partial least squares regression (PLSR) and the predicted values were used in visualization analysis. Finally, the possibility of prediction of chromium content in soil with hyperspectral data preprocessed by fractional differential in coalmine area was discussed. The results showed that fractional order differential model of the raw reflectance and the absorption rate transform both achieved the best performance at the 1.8-order derivative. Among all of the models through fractional order differential preprocessing, the model based on 1.8-order derivative of absorbance transform (RMSEC was 7.68mg/kg, R2c=0.83, RMSEP was 8.39mg/kg, R2p=0.78,RPD was 2.14) was much better than others, and had better performance in predicting Cr content in desert soil. Then the spatial distribution of the actual Cr content and its estimation values in soil of the study area were obtained by inverse distance weighted (IDW) algorithm. Moreover, the spatial distributions showed the same trend. The results showed that quantitative inversion of soil Cr content and the spatial distribution of large scale were feasible by this method. This research would provide scientific basis and technical support for the application in monitoring heavy metal contamination by hyperspectral data.