基于近红外光谱波长优选的土壤有机质含量预测研究
2018-11-13张小鸣汤宁
张小鸣 汤宁



摘 要: 近红外光谱技术是检测土壤信息的有效工具,为了提高预测模型的准确度和建模效率,需要对波长进行优选。提出SiPLS?GA?SPA特征波长提取方法,即协同区间偏最小二乘算法(SiPLS)、遗传算法(GA)和连续投影算法(SPA)对土壤有机质特征波长进行梯度提取,最终从1 050个波长中提取9个土壤有机质的特征波长。利用偏最小二乘回归(PLSR)和支持向量机回归(SVMR)建立6种基于特征波长的土壤有机质含量预测模型。结果表明:SiPLS?GA?SPA?SVMR模型的预测结果为RMSEP=1.15,R2=0.91,优于其他模型;SiPLS?GA?SPA特征波长提取方法能够简化预测模型,提高模型预测精度,为开发便携式近红外光谱土壤养分检测仪提供理论基础。
关键词: 近红外光谱; 特征波长; 协同区间偏最小二乘; 遗传算法; 连续投影算法; 支持向量机回归
中图分类号: TN929?34 文献标识码: A 文章编号: 1004?373X(2018)22?0126?04
Abstract: The near infrared spectroscopy technology is an effective tool for detecting soil information, and wavelength optimization is necessary to improve the accuracy and modeling efficiency of the prediction model. Therefore, an SiPLS?GA?SPA feature wavelength extraction method is proposed. The synergy interval partial least squares (SiPLS), genetic algorithm (GA) and successive projection algorithm (SPA) are combined to conduct gradient extraction for feature wavelengths of soil′s organic matter, and 9 feature wavelengths of soil′s organic matter are extracted from 1050 wavelengths. The partial least squares regression (PLSR) and support vector machine regression (SVMR) are adopted to establish 6 soil′s organic matter content prediction models based on feature wavelengths. The results show that, the prediction results of the SiPLS?GA?SPA?SVMR model (RMSEP=1.15, R2=0.91) are superior to other models, and the SiPLS?GA?SPA feature wavelength extraction method can simplify the prediction model and improve the prediction accuracy of the model, which provides a theoretical basis for the development of the portable near infrared spectroscopy soil nutrient detector.
Keywords: near infrared spectrum; feature wavelength; synergy interval partial least squares; genetic algorithm; successive projection algorithm; support vector machine regression
可見/近红外光谱区几乎包含了有机物中所有含氢基团的信息,信息量极为丰富,能够实现对物质的定性和定量分析[1]。光谱检测技术具有分析速度快、多组分同时测定、非破坏性分析、低分析成本和操作简单等显著特点[2]。能够适应现代精准农业(Precision Agriculture)对土壤养分信息实时准确、快速、大范围获取的要求。土壤有机质(SOM)是评价土壤肥力高低的重要指标,快速有效测定土壤有机质含量,对指导农业耕种中作物种植和施水施肥都有很大指导意义。
本文提出协同区间偏最小二乘(SiPLS)、遗传算法(GA)和连续投影算法(SPA)对土壤有机质的特征波长进行梯度提取,从1 050个波长中……
