基于高光谱的土壤有机质快速检测模型构建

Construction of Soil Organic Matter Rapid Detection Model Based on Hyperspectral

土壤有机质作为土壤肥力的重要指标。为实现对农田土壤有机质含量的快速获取,以山东省青岛市平度地区农田116个土壤样本为试验材料,利用ASD Field4地物光谱仪获取土壤光谱反射率,分析农田土壤的光谱反射特征,研究光谱反射率与定量化学方法测定有机质含量的相关关系,构建土壤有机质快速检测模型。所得高光谱数据结合(Savitzky-golay,SG)平滑算法,原始光谱曲线的一阶微分、对数的倒数和对数的倒数一阶微分3种变换方式对光谱数据进行预处理,通过相关系数法选取土壤有机质含量的敏感波段,分别建立多元线性回归(MLR)、BP神经网络(BPNN)和偏最小二乘回归(PLSR)模型,并对模型精度进行验证。结果表明,建立的MLR、BPNN和PLSR回归模型中,以BPNN模型精度最优,其建模样本集R2为0.7362,RMSE为4.6005,RPD为1.8550;验证集模型的R2为0.8086,RMSE为3.7772,RPD为2.2630。

Soil organic matter is an important index of soil fertility. In order to obtain farmland soil organic matter quickly,116 soil samples from Pingdu area of Qingdao City, Shandong Province were used as experimental materials. The spectral reflectance of soil was obtained by field4 spectrometer, and the spectral reflectance characteristics of farmland soil were analyzed. The correlation between spectral reflectance and quantitative chemical method to determine the content of organic matter was studied, and the rapid detection model of soil organic matter was constructed. The hyperspectral data were pretreated with savitzky Golay(SG) smoothing algorithm, first-order differential, reciprocal logarithm and reciprocal logarithm first-order differential of the original spectral curve. The sensitive bands of soil organic matter content were selected by correlation coefficient method, and multiple linear regression(MLR), BP neural network(BPNN) and partial least squares(PLS) were established respectively Regression(PLSR) model was used to verify the accuracy of the model.The results show that among the MLR, BPNN and PLSR regression models, BPNN model has the best accuracy, with R2 of0.7362, RMSE of 4.6005 and RPD of 1.8550 for the sample set, and R2 of 0.8086, RMSE of 3.7772 and RPD of 2.2630 for the validation set.