基于非负矩阵分解法的抗水分干扰土壤有机质高光谱估算  

Hyperspectral estimation of soil organic matter with anti-moisture interference based on Non-negative Matrix Factor

作　　者：丁仕杰 宋海燕[1] 张智峰 韩小平[1] Ding Shijie;Song Haiyan;Zhang Zhifeng;Han Xiaoping(College of Engineering,Shanxi Agricultural University,Taigu 030801,China)

机构地区：[1]山西农业大学工学院

出　　处：《山西农业大学学报（自然科学版）》2020年第1期102-109,共8页Journal of Shanxi Agricultural University(Natural Science Edition)

基　　金：国家重点研发计划(2018YFD0700300)

摘　　要：[目的]高光谱技术被普遍应用于土壤有机质的检测,而土壤水分在近红外波段中的吸收特性对土壤有机质的检测有很强干扰,消除水分对土壤有机质检测影响是提高模型预测精度的关键,开展土壤水分光谱影响及其消除方法的研究具有重要意义。[方法]本研究获取了山西省晋中市太谷县内不同区域的50个土壤样本在5种含水率(干土、5%、10%、15%和17%)下共250条高光谱曲线,用非负矩阵分解(Nonnegative matrix factor,NMF)对光谱分解重构,以去除水分对土壤有机质检测的影响。采用偏最小二乘(Partial least-squares,PLS)建立的土壤有机质定量预测模型对重构前后湿土土样的有机质含量进行了预测。[结果]经NMF分解重构后的湿样吸光度谱图与对应的干样相近;重构后的光谱数据对有机质预测相关系数(R)较湿样提高了0.059,预测标准差(Predicted standard deviation,SEP)降低了0.154,均方根误差(Root-mean-square error,RMSEP)降低了0.718。[结论]NMF能够在很大程度上削弱土壤水分对有机质检测的影响,提高湿土土样有机质含量的高光谱预测精度。[Objective]The near-infrared hyperspectroscopy technique is widely used in the detection of soil organic matter,while the strong absorption features of soil water in the near infrared spectrum severely affect the detection of soil organic matter.Therefore,the key point to improve the model prediction accuracy is to eliminate the influence of the moisture content on the soil organic matter detection.The present study was aimed to investigate the soil water spectrum effect and develop a model to eliminate its effects.[Methods]In this study,a total of 250 near-infrared hyperspectra were obtained from 50 soil samples at five moisture contents(dry,5%,10%,15%,and 17%)which were collected from different regions of Taigu County in Shanxi Province.Non-negative matrix factor(NMF)was used for spectral decomposition and reconstruction to eliminate the influence of water on soil organic matter detection.A Partial least-squares(Partial least-squares,PLS)quantitative prediction model was established to predict the organic matter content of wet soil samples before and after reconstruction.[Results]The results showed that the absorbance spectrum of wet sample was close to that of dry sample after NMF decomposition and reconstruction.Compared with wet samples,the correlation coefficient(R)of organic matter prediction was increased by 0.059 with the reconstructed spectral data,while the prediction standard deviation(SEP)and the root-mean-square error(RMSEP)were decreased by 0.154 and 0.718,respectively.[Conclusion]NMF can greatly weaken the influence of soil water on the organic matter detection,and improve the accuracy of hyperspectral prediction of organic matter content in wet soil samples.

关 键 词：红外高光谱 土壤有机质 非负矩阵分解 抗水分干扰 偏最小二乘 

分 类 号：S152.3[农业科学—土壤学] O657.3[农业科学—农业基础科学]