首页 | 本学科首页   官方微博 | 高级检索  
     检索      

土壤速效磷近红外迁移学习预测方法研究
引用本文:郑文瑞,李绍稳,韩亚鲁,石胜群,朱先志,金秀.土壤速效磷近红外迁移学习预测方法研究[J].分析测试学报,2020,39(10):1274-1281.
作者姓名:郑文瑞  李绍稳  韩亚鲁  石胜群  朱先志  金秀
作者单位:安徽农业大学信息与计算机学院智慧农业技术与装备安徽省重点实验室
基金项目:农业部“948”项目(2015-Z44,2016-X34);安徽省教育厅课题资助(KJ2019A0212)
摘    要:可见-近红外光谱技术是对土壤速效磷含量定量估测的有效手段,但某一地区土壤采集的光谱数据建立的模型在给其它地区使用时会出现预测精度低、模型失效等问题。该文以皖南土壤样本为源域,皖北土壤样本为目标域,通过迁移学习方法建立了预测模型,以提高土壤速效磷预测模型的准确性,并比较了迁移前后预测模型的精度。结果显示,皖南地区模型不能直接用于皖北地区,会出现模型失效问题,该模型的决定系数(R~2)和相对分析误差(RPD)分别为-0.19和0.92,预测均方根误差(RMSEP)为1.04;样本量不大的皖北地区建立模型的预测精度不高,R~2和RPD分别为0.61和1.60,RMSEP为0.60;而基于迁移成分分析(TCA)并加入部分皖北样本建立模型,可显著提高对皖北样本的预测精度,模型的R~2和RPD分别提升至0.79和2.18,RMSEP降低至0.44。表明基于TCA的方法能将皖南土壤速效磷预测模型应用于皖北,可提高皖北土壤速效磷预测模型准确性并降低成本,为土壤速效磷预测模型的广泛应用提供了新思路。

关 键 词:迁移学习  迁移成分分析  土壤速效磷  可见-近红外光谱

Study on Transfer Learning Prediction Methods for Soil Available Phosphorus NIR
ZHENG Wen-rui,LI Shao-wen,HAN Ya-lu,SHI Sheng-qun,ZHU Xian-zhi,JIN Xiu.Study on Transfer Learning Prediction Methods for Soil Available Phosphorus NIR[J].Journal of Instrumental Analysis,2020,39(10):1274-1281.
Authors:ZHENG Wen-rui  LI Shao-wen  HAN Ya-lu  SHI Sheng-qun  ZHU Xian-zhi  JIN Xiu
Institution:Anhui Agricultural University,school of information &computer,Anhui Provincial Key Laboratory of Smart Agricultural Technology and Equipment
Abstract:Visible and near infrared reflectance spectroscopy is an effective method for the quantitative estimation of soil available phosphorus content.However,when the spectral data collected from the soil of one region are used for those of other regions,problems such as low prediction accuracy and model failure may occur.In order to solve these problems,a prediction model through transfer learning method was established in this paper,with soil samples from southern Anhui province as the source domain and soil samples from northern Anhui province as the target domain to improve the accuracy of soil available phosphorus model prediction.By comparing the prediction accuracy of the model before and after the transfer,it was found that the model for southern Anhui region could not be directly used for Northern Anhui region,in that case,model failure may occur.The coefficient of determination(R2) and the ratio of prediction to deviation(RPD) of the model were-0.19 and 0.92,respectively,and the root mean square error of prediction(RMSEP) of the model was 1.04.The predictive accuracy of modeling for the North Anhui region with a small sample size was low,with R2 and RPD of 0.61 and 1.60,respectively,and RMSEP of 0.60.Based on the transfer component analysis(TCA) and adding some samples from northern Anhui to build the model,the prediction accuracy of the model for available phosphorus in samples from northern Anhui was significantly improved,with the R2 and RPD improved to 0.79 and 2.18,respectively,and the RMSEP reduced to 0.44.Results showed that the prediction model based on TCA method for soil available phosphorus in southern Anhui region could be applied to that in northern Anhui region,improving the prediction accuracy and reducing the modeling cost,and it provides a new idea for the wide application of prediction model for soil available phosphorus.
Keywords:transfer learning  transfer component analysis  soil available phosphorus  visible and near infrared reflectance spectroscopy
本文献已被 CNKI 等数据库收录!
点击此处可从《分析测试学报》浏览原始摘要信息
点击此处可从《分析测试学报》下载免费的PDF全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号