论文部分内容阅读
应用可见/近红外光谱技术对土壤有机质含量进行了定量分析和预测,为土壤肥力快速测定和评价提供依据.利用ASD FieldSpec 3 Hi-Res光谱仪对116份不同有机质含量的土壤样本进行光谱测量,系统分析了土壤有机质含量与350～2500 nm波段范围光谱反射率之间的关系.利用PLS和小波-BP神经网络对350～2500 nm整个波段范围和剔除水波段的光谱数据进行分析.两种建模方法的结果均表明剔除水波段的预测效果较好,其中,PLS模型预测的相关系数R为0.8416,均方根误差RMSEP为0.2848,相对分析误差RPD为1.7768,WT-BP神经网络模型预测的R为0.9167,RMSEP为0.2196,RPD为2.3043.预测结果表明,PLS模型可以对土壤有机质含量进行粗略估测,而BP神经网络可实现较精确的预测.
The soil organic matter content was quantitatively analyzed and predicted by visible / near infrared spectroscopy (NIRS), which provided the basis for rapid soil fertility determination and evaluation.116 soil samples with different organic matter content were measured by ASD FieldSpec 3 Hi-Res spectrometer, and the system The relationship between soil organic matter content and spectral reflectance in the range of 350-2500 nm was analyzed.The spectral data of the entire band range of 350-2500 nm and the water band excluded were analyzed by PLS and wavelet-BP neural networks.The two models The results show that the prediction of cull water band is better. The correlation coefficient R predicted by PLS model is 0.8416, the root mean square error RMSEP is 0.2848 and the relative analysis error RPD is 1.7768. The prediction of R Which is 0.9167, RMSEP is 0.2196 and RPD is 2.3043. Prediction results show that PLS model can roughly estimate soil organic matter content, while BP neural network can achieve more accurate prediction.