摘要
以新疆阜康不同程度人类干扰下的土壤为研究对象,探讨高精度的硝态氮含量的反演模型。测量土壤样品的室外高光谱数据和通过化学分析获取硝态氮含量,并对原始土壤高光谱进行一阶导数和二阶导数变换,以通过0.01显著性水平的波段作为敏感波段,利用BP神经网络和逐步多元回归模型(SMLR)建立8个硝态氮含量的定量估测模型。仿真结果表明:与SMLR相比,BP模型能显著提升2种土壤中的硝态氮含量的反演精度。尤其是原始高光谱数据经过一阶导数变换后的BP模型精度最高,能对硝态氮含量进行精确的预测,其在无人类干扰区域的土壤测试集中的相对分析误差(RPD)为2.884,决定系数(R2)为0.874;在有人类干扰区域的土壤测试集中的RPD为2.226,R2为0.929。SMLR模型在2种土壤中的RPD值均在1.1左右,说明基于SMLR模型对硝态氮含量的预测能力很差。
By taking the soil disturbed by different degree human beings activities in Fukang of Xinjiang Uygur Autonomous Region as the research object, this paper discusses the inversion model of high-precision nitrate nitrogen content. The outdoor hyperspectral data of soil samples are measured and the nitrate nitrogen content is obtained by chemical analysis. The first derivative and the second derivative of the original hyperspectral data are transformed. By taking the 0.01 significant band as the sensitive band, the quantitative estimation models of 8 nitrate nitrogen contents are established by using BP neural network and stepwise multiple regression(SMLR) model. The simulation results show that compared with SMLR model, BP model can significantly improve the inversion accuracy of nitrate nitrogen content in two soils. In particular, the BP model of the original hyperspectral data after the first derivative transformation has the highest accuracy, which can accurately predict the nitrate nitrogen content. The relative analysis error(RPD) of the soil test in the area without human disturbance is 2.884 and the determination coefficient(R2)is 0.874. The RPD and R2 of soil test in the area with human disturbance are 2.226 and 0.929 respectively. The RPD value of SMLR model in two kinds of soil is about 1.1, which shows that the prediction ability of SMLR to nitrate nitrogen content is very poor.
作者
田安红
杨丽华
崔丽梅
付承彪
赵俊三
熊黑钢
于龙
TIAN Anhong;YANG Lihua;CUI Limei;FU Chengbiao;ZHAO Junsan;XIONG Heigang;YU Long(College of Information Engineering,Qujing Normal University,Qujing 655011,China;Faculty of Land Resource Engineering,Kunming University of Science and Technology,Kunming 650093,China;College of Applied Arts and Science,Beijing Union University,Beijing 100083,China;College of City,Qujing Normal University,Qujing 655011,China)
出处
《实验技术与管理》
CAS
北大核心
2020年第4期51-56,共6页
Experimental Technology and Management
基金
国家自然科学基金项目(41901065,41671198,41761081,31660680)
教育部产学合作协同育人项目(201802156014)
曲靖师范学院教师教育研究专项项目(2019JZ001)。
关键词
不同程度人类干扰土壤
硝态氮含量
高光谱
BP神经网络模型
different degree of human activities
soils
nitrate nitrogen content
hyperspectral
BP neural network model
