摘要
应用近红外光谱和中红外光谱对浙江省衢州红壤和海宁青紫泥2种典型土壤的氮(N)、磷(P)和钾(K)等养分进行快速测试;试验共采集80个样本,其中60个用于建模,20个用于预测;在获取光谱信息的基础上,分别采用偏最小二乘-支持向量机(PLS-LS-SVM)和偏最小二乘-人工神经网络(PLS-BP/ANN)2种方法进行建模.结果表明:2种模型的预测结果均比较理想,在小样本的学习预测上,PLS-LS-SVM比PLS-BP/ANN更精确一些;近红外光谱和中红外光谱2个波段对N含量的预测效果均较好,PLS-LS-SVM模型的预测相关系数分别为0.876和0.867;中红外波段对P和K的预测效果更好,PLS-LS-SVM模型的预测相关系数分别为0.938(P)和0.803(K).这为土壤养分的快速测试提供了一种新方法.
The nutritional parameters(N,P and K) in two typical soils(red soil in Quzhou and purplish clayey soil in Haining) in Zhejiang Province were determined using near infrared(NIR) and middle infrared(MIR) spectroscopy.A total of 80 soil samples were collected,60(30 for each variety) samples of which were used as calibration set,and the remaining 20 samples were used as validation set.After spectral scanning,partial least squares-least squares-support vector machine(PLS-LS-SVM) and partial least squares-back propagation neural networks(PLS-BP/ANN) were applied to develop the calibration models.The results indicated that both PLS-LS-SVM and PLS-BP/ANN achieved good prediction results,and PLS-LS-SVM were more suitable for small soil samples,both NIR and MIR achieved good prediction results for N detection by PLS-LS-SVM model with correlation coefficients r=0.876 and r=0.867,respectively.The MIR was better than NIR for the prediction of P and K,and the best results were obtained by PLS-LS-SVM model with r=0.938 for P and r=0.803 for K.It supplies a new way for the fast and accurate detection of nutritional parameters in soil.
出处
《浙江大学学报(农业与生命科学版)》
CAS
CSCD
北大核心
2010年第4期445-450,共6页
Journal of Zhejiang University:Agriculture and Life Sciences
基金
国家高技术研究发展计划863"资助项目(2007AA10Z210)
国家农业科技成果转化基金资助项目(2009GB23600517)
科技型中小企业技术创新基金资助项目(09C26213303994)
关键词
光谱技术
土壤
养分
人工神经网络
支持向量机
偏最小二乘法
spectroscopy
soil
nutrition
artificial neural networks
support vector machines
partial least squares analysis