论文部分内容阅读
为建立一种能够同时快速检测土壤全磷和全钾的定量估计模型,该文采用近红外漫反射技术对赣南脐橙果园的土壤进行研究,对56个土样风干、过筛,然后进行光谱采集和化学分析。光谱经过Savitzky-Golay平滑后再用一阶微分变换的方法进行预处理,分别应用偏最小二乘回归(partial least square regress PLS)、主成分回归(principal component regression PCR)和最小二乘支持向量机(least squares support vector machine LS-SVM)3种方法,在4 000~7 500 cm~(-1)波数范围内,建立赣南脐橙果园土壤全磷和全钾快速定量检测模型。结果发现在建立土壤全磷模型时,PLS和PCR的预测模型效果均不理想,但LS-SVM建立的模型较为理想,其预测相关系数(correlation coefficient of prediction R_P)为0.884,预测集均方根误差(the root mean square error of prediction RMSEP)为0.341,预测相对分析误差(residual predictive deviation RPD)为2.59。在建立土壤全钾模型时,PLS、PCR和LS-SVM建立3种模型效果均理想,其中以LS-SVM模型最理想,其预测相关系数(R_P)为0.971,预测集均方根误差(RMSEP)为0.714,预测相对分析误差(RPD)为5.12。研究表明,采用LS-SVM建立的土壤全磷和全钾模型对实现土壤全磷和全钾含量快速检测具有可行性。
In order to establish a quantitative estimation model that can simultaneously detect total phosphorus and total potassium in soil simultaneously, this paper studied the soil of Gannan Navel orange orchard by near-infrared diffuse reflectance spectroscopy. 56 soil samples were air-dried and sieved, Acquisition and chemical analysis. The spectra were pre-processed by Savitzky-Golay smoothing and then first-order differential transformation. Partial least square regress (PLS), principal component regression (PCR) and least-squares support vector machine (LS-SVM) method was used to establish a rapid and quantitative model of total phosphorus and total potassium in soil of Gannan Navel orange orchards in the range of 4 000-7 500 cm -1. The results showed that the PLS and PCR prediction models were not effective in establishing the soil total phosphorus model, but the LS-SVM model was ideal, the correlation coefficient of prediction R_P was 0.884, the root mean square of prediction set The root mean square error of prediction (RMSEP) was 0.341 and the residual predictive deviation RPD was 2.59. In the establishment of soil total potassium model, PLS, PCR and LS-SVM established three models with satisfactory results. Among them, LS-SVM model was the most ideal model with the prediction correlation coefficient (R_P) of 0.971 and the root mean square error of prediction set ) Was 0.714, with a predicted relative analysis error (RPD) of 5.12. The results show that the soil total phosphorus and total potassium model established by LS-SVM is feasible for rapid determination of total phosphorus and total potassium in soil.