文章摘要
基于决策树模型的耕地地力与玉米丝黑穗病发生关系研究
A study of relationships between cultivated land fertility and corn head smut using a decision tree model
投稿时间:2021-04-02  
DOI:10.13254/j.jare.2021.0193
中文关键词: 玉米丝黑穗病,耕地地力,分类与回归树模型,随机森林模型,极端随机树模型
英文关键词: corn head smut, cultivated land fertility, classification and regression tree model, random forest model, extremely randomized trees model
基金项目:国家重点研发计划项目(2017YFD0200607);中国农业科学院创新工程项目(2021CX013)
作者单位E-mail
陈丽 中国农业科学院农业信息研究所, 北京 100081
农业农村部农业大数据重点实验室, 北京 100081 
 
崔运鹏 中国农业科学院农业信息研究所, 北京 100081
农业农村部农业大数据重点实验室, 北京 100081 
cuiyunpeng@caas.cn 
王末 中国农业科学院农业信息研究所, 北京 100081
农业农村部农业大数据重点实验室, 北京 100081 
 
牛永春 中国农业科学院农业资源与农业区划研究所, 北京 100081)  
徐爱国 中国农业科学院农业资源与农业区划研究所, 北京 100081)  
刘珂艺 中国农业科学院农业信息研究所, 北京 100081
农业农村部农业大数据重点实验室, 北京 100081 
 
刘娟 中国农业科学院农业信息研究所, 北京 100081
农业农村部农业大数据重点实验室, 北京 100081 
 
侯颖 中国农业科学院农业信息研究所, 北京 100081
农业农村部农业大数据重点实验室, 北京 100081 
 
摘要点击次数: 108
全文下载次数: 86
中文摘要:
      为了探究耕地地力与玉米丝黑穗病发生关系,本研究以515个主要玉米种植县域为研究区域,选取有机质(Organic matter,OM)、全氮(Total nitrogen,TN)、全磷(Total phosphorus,TP)、全钾(Total potassium,TK)、有效磷(Available phosphorus,AP)、速效钾(Available potassium,AK)和pH 7种耕地地力因子,分别利用分类与回归树(Classification and regression tree,CART)模型、随机森林(Random forest,RF)模型和极端随机树(Extremely randomized trees,ERT)模型构建了玉米丝黑穗病发生程度与耕地地力因子关系模型,并进行了3种模型效果比较。结果表明:RF和ERT模型总分类性能明显优于CART模型,3个模型在病害发生程度1级(GⅠ)上的查准率(Precision,Pr)、查全率(Recall,Re)、F1 score(F1)值均较高,分类效果比病害发生程度2级(GⅡ)要好,但考虑到准确监测病害高发情况、减少高发病情况在分类预测中漏分机率对开展病害防治的重要性,确定ERT模型为最佳优选分类器。耕地地力特征变量与病害发生程度重要性分析表明,玉米丝黑穗病发生程度与耕地地力因子AP、TK、pH和TP具有一定的相关性。研究结果为深入探究耕地地力对玉米丝黑穗病影响机理提供了线索和支撑。
英文摘要:
      To explore the relationship between cultivated land fertility and corn head smut occurrence, 515 main corn growing counties were selected, and the relationship between corn head smut incidence and cultivated land fertility was modeled using classification and regression tree(CART), random forest(RF) and extreme randomized trees(ERT) models, with organic matter(OM), total nitrogen(TN), total phosphorus(TP), total potassium(TK), available phosphorus(AP), available potassium(AK), and PH as characteristic variables. The classification accuracy of the different models were compared. The comprehensive classification performance of RF and ERT was obviously better than that of CART. The three models had higher Precision(Pr), Recall (Re), and F1 score(F1) values on disease grade 1 (GⅠ), and the classification effect was better than that of disease grade 2(G Ⅱ). Considering the importance of accurately monitoring the high incidence of diseases and reducing leakage rates in classification and prediction for disease prevention and control, the ERT model, with its higher Re value on G Ⅱ was finally determined to be the best classifier. Additionally, importance analysis between the variables characteristic of cultivated land fertility and degree of disease occurrence showed that the incidence of corn head smut correlated appreciably with AP, TK, pH, and TP. The results provide clues and support for further mechanistic research into effects of cultivated land fertility factors on corn head smut.
HTML   查看全文   查看/发表评论  下载PDF阅读器
关闭