引用本文:李怡春,王昌昆,潘恺,刘娅,吴士文,刘杰,徐爱爱,张芳芳,潘贤章.基于PLSR的土壤颜色预测方法及其与色系转换法的对比研究[J].土壤学报,2018,55(6):1411-1421.
LI Yichun,WANG Changkun,PAN Kai,LIU Ya,WU Shiwen,LIU Jie,XU AIai,ZHANG Fangfang,PAN Xianzhang.PLSR-Based Prediction of Soil Color and Its Comparison with Color Space Conversion Method[J].Acta Pedologica Sinica,2018,55(6):1411-1421
【打印本页】   【HTML】   【下载PDF全文】   查看/发表评论  【EndNote】   【RefMan】   【BibTex】
←前一篇|后一篇→ 过刊浏览    高级检索
本文已被:浏览 156次   下载 238 本文二维码信息
码上扫一扫!
分享到: 微信 更多
基于PLSR的土壤颜色预测方法及其与色系转换法的对比研究
李怡春,王昌昆,潘恺,刘娅,吴士文,刘杰,徐爱爱,张芳芳,潘贤章
中国科学院南京土壤研究所,中国科学院南京土壤研究所,中国科学院南京土壤研究所,中国科学院南京土壤研究所,中国科学院南京土壤研究所,中国科学院南京土壤研究所,中国科学院南京土壤研究所,中国科学院南京土壤研究所,中国科学院南京土壤研究所
摘要:
传统的土壤颜色测定主要采用蒙塞尔比色卡比对,精度高,但费时费力。近年来尝试采用色系转换法预测土壤颜色,方法较为简便。基于土壤高光谱反射率和偏最小二乘回归(PLSR)方法进行土壤颜色预测,并与色系转换法进行对比研究。采集皖赣鄂交界地区76个不同颜色的土壤样品,分别采用PLSR及色系转换法进行土壤颜色预测,并与实测结果进行了对比。结果表明,PLSR交叉验证的Rcv2分别达到0.62、0.61和0.75,测定值标准偏差与标准预测误差的比值(RPD)分别达到1.94、1.67和2.15,说明PLSR模型用于土壤颜色预测是可行的;其均方根误差(RMSE)仅为1.32、0.55和0.97个单位,较色系转换法的RMSE分别低0.94、1.24 和0.95个单位,其HV/C整体预测误差∆E的平均值为1.91,较色系转换法的平均值低5.16,说明PLSR方法预测土壤蒙塞尔颜色较色系转换法更优。该方法为土壤颜色的获取提供了一种新的途径。
关键词:  土壤颜色  色系转换  光谱  偏最小二乘回归
DOI:10.11766/trxb201803300469
分类号:
基金项目:科技部基础性专项课题(2015FY110700S5)、STS项目(KFJ-SW-STS-168)
PLSR-Based Prediction of Soil Color and Its Comparison with Color Space Conversion Method
liyichun,WANG Changkun,PAN Kai,LIU Ya,WU Shiwen,LIU Jie,XU AIai,ZHANG Fangfang and PAN Xianzhang
Institute of Soil Science,Chinese Academy of Sciences,Institute of Soil Science,Chinese Academy of Sciences,Institute of Soil Science,Chinese Academy of Sciences,Institute of Soil Science,Chinese Academy of Sciences,Institute of Soil Science,Chinese Academy of Sciences,Institute of Soil Science,Chinese Academy of Sciences,Institute of Soil Science,Chinese Academy of Sciences,Institute of Soil Science,Chinese Academy of Sciences,Institute of Soil Science,Chinese Academy of Sciences
Abstract:
The color of a soil may, to a certain extent, reflect degree in development, type and fertility of the soil. Traditionally, soil color is measured with the Munsell colorimetry, which, though quite high in accuracy is time-consuming and low in efficiency. It is, therefore, essential to explore for a quick and accurate method to measure soil colors. Nowadays remote sensing and proximal sensing methods can be used to obtain soil information, and numerous attempts have been made to extract soil color information from soil spectra. For that end, color space conversion (CSC) method is a commonly used one. It uses mathematical formulas to match colors between different coordinates, so as to realize prediction of soil colors. The first step of this method is to extract average reflectance values of the RGB bands from spectral reflectance and then converts them into XYZ values in the CIE XYZ coordinate, and further into HV/C values in the munsell coordinate. In this paper, a novel method was introduced to predict soil colors using partial least squares regression (PLSR) of hyperspectral reflectance of soils, and then comparison was made between PLSR and CSC in prediction accuracy. 【Method】A total of 76 soil samples different in colors were collected in the bordering area of Anhui, Jiangxi and Hubei Province for the study, covering soil types e.g. red soil (Argi-Udic Ferrosols), paddy soil (Fe-accumuli-Stagnic Anthrosols), yellow-brown soil (Ferri-Udic Argosols), fluvo-aquic soil (Ochri-Aquic Cambosols), purple soil (Dystric Purpli-Udic Cambosols) and yellow soil (Ali-Perudic Argosols) in the study area. After being air-dried, the soil samples were determined in color through color matching with the Munsell color system, and their spectral reflectance was acquired simultaneously with the aid of the ASD spectrometer. Then PLSR and CSC was applied separately to predict colors of the soil samples. 【Result】Results show that the PLSR model can be well used to predict soil Hue (H), Value (V), and Chroma (C) with cross validation coefficient Rcv2) being 0.62, 0.61 and 0.75 respectively, and RPD being 1.94, 1.67 and 2.15 respectively, which suggests that it is feasible to use the PLSR method to predict soil colors and that the mean square root error (RMSE) of H, V and C predicted with PLSR was only 1.32, 0.55 and 0.97 units, respectively, and 0.94, 1.24 and 0.95 lower than their respective ones predicted with the CSC method. The former, being 1.91, was 5.16 lower than the latter in mean ∆E, the mean HV/C comprehensive index. Analysis of reasons for that reveals that PLSR uses the spectral reflectance information of all the bands, while CSC makes use of mean reflectance of Red, Green and Blue bands only. Furthermore, certain errors inevitably occur in every step of the conversion of CSC.【Conclusion】Therefore, it could be concluded that the PLSR method is superior to the CSC method in predicting Munsell color of a soil. And compared the conventional soil color measuring methods, this one saves time and labor by a large mirgin. So this method opens up a new way for quick soil color acquisition via soil spectrum.
Key words:  Soil munsell color  Color space conversion method  Spectroscopy  Partial least squares regression