摘要
为解决在土壤速效钾含量的高光谱定量预测分析过程中,光谱数据维数高、冗余度较大等问题,提出了一种结合K均值算法(K-means)和连续投影算法(SPA)的高光谱特征波段选择方法。该算法首先将全波段数据分别根据不同的距离度量进行K-means聚类分析,之后对聚类后的每个波段簇分别使用SPA法提取其中的特征波段。对全波段组合、传统SPA法提取的特征波段组合以及结合K-means聚类与SPA法提取的特征波段组合分别建立土壤速效钾含量的BP神经网络预测模型,通过对比模型预测效果来比较特征波段选择方法的性能。以盐城市348份土壤样品进行试验,结果表明,结合K均值算法与连续投影算法的特征波段选择方法可以有效地解决光谱预测分析过程中的数据冗余问题,实现对土壤速效钾含量快速精确预测分析。
In order to solve the problems of high dimensionality and redundancy of hyperspectral spectral data in hyperspectral quantitative prediction and analysis of soil available potassium content,a hyperspectral band selection method based on K-means algorithm and successive projection algorithm(SPA)was proposed.Firstly,the full-band data were clustered by K-means based on different distance measures,and then the characteristic bands were extracted by SPA method for each band cluster after clustering.BP neural network prediction models of soil available potassium content were established for full-band combination,combination of characteristic bands extracted by traditional SPA method and combination of characteristic bands extracted by K-means clustering and SPA,respectively.The performance of characteristic band selection methods was evaluated by comparing the prediction effects of the models.The 348 soil samples from Yancheng City were experimented.The characteristic band selection method based on K-means algorithm and successive projection algorithm can effectively solve the problem of data redundancy in the process of spectral prediction and analysis,and achieve the rapid and accurate prediction and analysis of soil available potassium content.
作者
朱淑鑫
杨宸
顾兴健
张永春
艾玉春
徐焕良
ZHU Shu-xin;YANG Chen;GU Xing-jian;ZHANG Yong-chun;AI Yu-chun;XU Huan-liang(College of Information Technology,Nanjing Agricultural University,Nanjing 210095,China;Institute of Agricultural Resources and Environment,Jiangsu Academy of Agricultural Sciences,Nanjing 210014,China)
出处
《江苏农业学报》
CSCD
北大核心
2020年第2期358-365,共8页
Jiangsu Journal of Agricultural Sciences
基金
中央高校基本科研业务费专项资金项目(KYGL201808)
江苏省农业科技自主创新基金项目[CX(17)1001]。
作者简介
朱淑鑫(1978-),女,江苏盐城人,硕士,副教授,主要从事农业信息化与土壤数据研究,(E-mail)zsx@njau.edu.cn;通讯作者:徐焕良,(E-mail)huanliangxu@njau.edu.cn。
