【摘要】 采用高光谱技术获得的数据进行土壤有机质含量的反演和估测是近年来的研究热点。为确定有效的估测建模方法,利用地面实测的土壤高光谱反射率及有机质含量等数据,采用小波分析方法实现去噪,包络线去除法实现建模参数提取和数据量压缩,结合多种不同的数据变换方法,利用BP神经网络法、多元线性回归法及最小二乘回归法建立不同的估测模型。对比发现,BP神经网络模型的估测效果优于回归模型,其中结合对数的平方变换和神经网络所建立的模型为最优估测模型,模型的决定系数达到0. 933,检验样本的均方根误差达到0. 069。实验证明,BP神经网络+对数的平方变换模型的学习机制适用于土壤有机质含量地面高光谱估测且效果好。通过在建模因子层面上进行数据变换建立了较好的估测模型,其研究方法、模型和结论,对土壤有机质含量地面高光谱估测具有一定的参考意义。更多还原

【Abstract】 Using the data obtained by hyperspectral techniques to estimate the content of soil organic matter is a hotspot in recent years.For the purpose of determining the effective estimation modeling method,specific data such as reflectance obtained by hyperspectral on ground and organic matter content were used in this paper.Wavelet analysis was used to remove the noise,and continuum removal was used to extract the parameters and compress the data.Combining a variety of different data transformation methods and utilizing BP neural networks,multiple linear regression(MLR) and least squares regression(LSR),many different estimation models of soil were established.It is found that the neural network method is superior to the regression model among various data transformation methods after comparing different estimation models established by the three modeling methods.The optimal estimation model is the model established by the combination of logarithmic square transformation and neural network.The R2 of the model is 0.933 and the RMSE is 0.069.The authors creatively carried out the data transformation at the modeling factor level and established the good estimation model.It is shown that the learning mechanism of BP + LS model is suitable for hyperspectral estimation of soil organic matter and works well.The methods,models and conclusions of this paper have some reference significance for the hyperspectral estimation of soil organic matter.更多还原