摘要
针对浙江省4种典型土壤,研究应用可见-近红外光谱、近红外光谱和中红外光谱3个波段范围进行土壤快速分类的方法.在获取光谱信息的基础上,采用不同光谱建模方法以提高检测精度,简化分析计算;并分别采用主成分分析结合人工神经网络(PCA-ANN/BP)、偏最小二乘法(PLS)和偏最小二乘法结合人工神经网络(PLS-ANN)3种方法进行建模.结果表明:中红外光谱波段对土壤分类的效果不理想,而可见-近红外光谱、近红外光谱波段均能较好地进行土壤分类;在可见-近红外波段,PLS-ANN模型对土壤的分类效果优于PCA-ANN/BP和PLS模型,为土壤快速准确分类提供了一种简便可行的方法.
Three spectral regions were investigated for the discrimination of four main soil varieties in Zhejiang Province.The three spectral regions included the visible-near infrared (Vis-NIR),near infrared (NIR),middle-infrared (MIR) spectroscopy.After spectral collection,different calibration methods were applied to improve the detection precision and to reduce the computational complexity.Principal component analysis (PCA) combined with back propagation neural networks (ANN/BP),partial least squares (PLS) and PLS combined with ANN (PLS-ANN) were applied for calibration models.The results indicated that the performance of MIR was not quite acceptable.However,Vis-NIR and NIR made an excellent discrimination precision for soil varieties.In Vis-NIR region,PLS-ANN model outperformed PCA-ANN/BP and PLS models.This supplies a convenient and feasible approach for the rapid discrimination of soil variety.
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3期282-286,共5页
Journal of Zhejiang University:Agriculture and Life Sciences
基金
"十一五"国家科技支撑计划资助项目(2006BAD10A0902)
国家农业科技成果转化基金资助项目(2009GB23600517)
浙江省宁波市自然科学基金资助项目(2007A610080)
关键词
光谱技术
土壤
分类
人工神经网络
主成分分析
偏最小二乘法
spectroscopy
soil
variety discrimination
artificial neural networks
principal component analysis
partial least squares analysis