棉花冠层氮素的高光谱监测与模型研究  

Hyperspectral Monitoring and Model Study of Nitrogen in Cotton Canopy

作　　者：李贞 王家强[1] 李紫琴 邹德秋 张小功 韩建文 柳维扬[1] LI Zhen;WANG Jiaqiang;LI Ziqin;ZOU Deqiu;ZHANG Xiaogong;HAN Jianwen;LIU Weiyang(College of Agriculture/Key Laboratory of Genetic Improvement and Efficient Production for Specialty Crops in Arid Southern Xinjiang of Xinjiang Corps,Tarim University,Alar,Xinjiang 843300,China)

机构地区：[1]塔里木大学农学院/南疆干旱区特色作物遗传改良与高效生产兵团重点实验室,新疆阿拉尔843300

出　　处：《江西农业大学学报》2023年第5期1273-1284,共12页Acta Agriculturae Universitatis Jiangxiensis

基　　金：新疆生产建设兵团科技创新人才计划项目(2022CB001-07);新疆生产建设兵团科技计划项目(2021DB019)。

摘　　要：【目的】构建棉花冠层氮素含量的最佳光谱估算模型,进行棉花氮素监测方法的研究,对快速识别棉花氮素含量变化特征,实现棉花氮素营养的快速、无损监测,基于遥感技术的棉花氮素营养诊断及指导棉田氮肥的科学精准施用意义重大。【方法】以盆栽棉花为研究材料,设置0 kg/hm^(2)(T0)、100 kg/hm^(2)(T1)、150 kg/hm^(2)(T2)、200 kg/hm^(2)(T3)、250kg/hm^(2)以上(T4)5个施氮梯度。将棉花冠层含氮量与SG平滑(savitzky-golay,SG)、连续小波变换(continuous wavelet transform,CWT)和SG+CWT 3种不同高光谱预处理方式的反射率数据相结合,采用反向传播神经网络(BP neural network,BPNN)和随机森林(random forest,RF)2种建模方法,对棉花冠层含氮量进行估算建模。【结果】(1)棉花冠层光谱反射率经过SG、CWT、SG+CWT处理后可以有效突出棉花叶片光谱特征,尤其是局部光谱吸收特征更加显著。(2)3种光谱预处理方法相比较,SG+CWT较好地提高了棉花光谱反射率与棉花冠层含氮量的相关性;(3)利用SG+CWT变换光谱建立的棉花氮素BPNN和RF估测模型的精度均通过了精度验证,其中RF模型的建模精度高于BPNN模型,其建模集的R2,RPD分别为0.848,2.084;验证集的R2,RPD分别为0.813,1.759。【结论】对棉花冠层反射光谱进行SG+CWT变换比单一进行SG平滑或CWT处理在棉花冠层氮素估算建模方面具有更大的优势。结果显示基于SG+CWT处理的棉花冠层光谱数据构建的RF模型是估算棉花氮素含量的最优模型,此结论为基于遥感技术的棉花氮素营养诊断及棉田的精确施肥管理提供了必要的技术支撑。[Objective]The optimal spectral estimation model of nitrogen content in cotton canopy was established.Studying cotton nitrogen monitoring methods is of great significance for rapidly identifying the change characteristics of cotton nitrogen content,realizing rapid and non-destructive monitoring of cotton nitrogen nutrition,and conducting cotton nitrogen nutrition diagnosis based on remote sensing technology and guiding the scientific and accurate application of nitrogen fertilizer in cotton fields.[Method]Potted cotton was used as the research material,5 nitrogen gradients,namely 0 kg/hm^(2)(T0),100 kg/hm^(2)(T1),150 kg/hm^(2)(T2),200 kg/hm^(2)(T3)and 250 kg/hm^(2)(T4)were designed.The nitrogen content in cotton canopy was combined with the reflectance data of three different hyperspectral preprocessing methods,Savitzky-Golay(SG),Continuous Wavelet Transform(CWT)and SG+CWT.Two modeling methods,BP Neural Network(BPNN)and Random Forest(RF),were used to estimate and model the nitrogen content in cotton canopy.[Result](1)The spectral reflectance of cotton canopy after SG,CWT,SG+CWT treatments could effectively highlight the spectral characteristics of cotton leaves,the local spectral absorption characteristics were especially more significant(.2)Among the three spectral preprocessing methods,SG+CWT improved effectively the correlation between cotton spectral reflectance and cotton canopy nitrogen content.(3)The accuracy of the BPNN and RF estimation models of cotton nitrogen established by SG+CWT transform spectrum has passed the accuracy verification.The accuracy of the RF model was higher than that of the BPNN model,and the R2 and RPD of the modeling set were 0.848 and 2.084,respectively.The R2 and RPD of the verification set were 0.813 and 1.759,respectively.[Conclusion]SG+CWT transformation of cotton canopy reflectance spectrum is better than single SG smoothing or CWT treatment in estimation modeling of cotton canopy nitrogen.The results showed that RF model based on SG+CWT cotton canopy spectral data was the be

关 键 词：棉花氮素 光谱预处理 连续小波变换 模型 

分 类 号：S127[农业科学—农业基础科学] S562