土壤有机质含量反演方法:土壤高光谱数据微分-瞬时频率变换

Soil Organic Matter Content Inversion Methods:Differential-Instantaneous Frequency Transform of Soil Hyperspectral Data

土壤有机质具有改善土壤结构、增强土壤肥力和保持水分的能力,有机质含量是衡量土壤肥力的重要指标。高光谱数据为有机质含量估算提供了一种快速、有效的解决途径。为了提升土壤有机质含量的估测精度,本文结合微分变换和希尔伯特变换,提出微分-瞬时频率变换方法,对土壤有机质含量进行精确估测。首先,利用微分变换具有特征增强等特点,对土壤光谱进行微分变换,并采用希尔伯特变换定义土壤微分光谱的解析信号,求其瞬时频率,从而更敏锐、深入地捕捉土壤微分光谱的局部特征;然后,采用相关性分析算法提取变换后光谱的敏感特征;最后,利用偏最小二乘(Partial Least Squares Regression,PLSR)和支持向量机(Support Vector Machine,SVM)2种算法构建土壤有机质含量估测模型。为了验证本文方法的有效性,将本文方法与传统光谱变换和瞬时频率等7种光谱处理算法建立的估算模型进行精度对比。实验结果表明:①微分-瞬时频率变换算法的测试精度为R2=0.85,RMSE=0.98,LCCC=0.84均高于其他算法,具有最佳的土壤有机质含量估测能力。②对比PLSR和SVM模型的建模效果,发现所有变换方法中SVM模型的建模精度和测试精度均优于PLSR模型。综上,基于微分-瞬时频率方法在土壤有机质含量反演中具有较高的估算精度,可为高光谱遥感技术在土壤有机质含量反演上提供新思路,进而服务于大尺度精准农业中的土壤监测。

Soil organic matter has the ability to improve soil structure,enhance soil fertility,and retain water.Soil organic matter content is an important index to measure soil fertility.Traditional methods based on physical and chemical analysis of soil organic matter content are often employed in laboratory.However,these methods are time-and labor-consuming,complex and expensive,and it is difficult to meet the practical and real-time needs of field fertilization management in precision agriculture.In contrast,hyperspectral data,with high spectral resolution,can provide more detailed and accurate surface soil information,which can meet various needs of agricultural production in a timely manner.To improve the estimation accuracy of soil organic matter content based on hyperspectral data,this paper combines differential transform and Hilbert transform,and proposes a differential-instantaneous frequency transform method to accurately estimate soil organic matter content.Firstly,the differential transform is used to transform soil spectrum,and the Hilbert transform is used to define the analytical signal of the soil differential spectrum.Thus,the instantaneous frequency is obtained,so as to capture the local characteristics of soil differential spectrum more accurately.Then,the correlation analysis is conducted to extract the sensitive features of the transformed spectrum.Finally,the estimation model of soil organic matter content is constructed using Partial Least Squares Regression(PLSR)and Support Vector Machine(SVM).In order to verify the validity of our proposed method,the accuracy of the proposed method is compared with that of other estimation models established using the first order differential transform,instantaneous frequency transform,first order differential transform of logarithm,and first order differential transform of reciprocal spectrum processing algorithms.The experimental results show that:(1)the estimation accuracy of the differential-instantaneous frequency transform algorithm(R2=0.85,RMSE=0.98,LCCC=0.84)is higher than that of other comparison algorithms,and thus has the best estimation ability of soil organic matter content;(2)By comparing the performance of PLSR and SVM models,it is found that the modeling accuracy and testing accuracy of SVM model are higher than those of PLSR model among all transformation methods.In summary,the differential-instantaneous frequency method has a high estimation accuracy in soil organic matter content inversion,which provides a new idea for soil organic matter content inversion from hyperspectral remote sensing and thereby improving soil monitoring in large-scale precision agriculture.