基于无人机多光谱和多波段组合纹理的马铃薯LAI估算  

Estimation of Potato LAI Using UAV Multispectral and Multiband Combined Textures

作　　者：郭立笑 陈志超[1] 马彦鹏 边明博 樊意广 陈日强 刘杨 冯海宽[2,3] GUO Li-xiao;CHEN Zhi-chao;MA Yan-peng;BIAN Ming-bo;FAN Yi-guang;CHEN Ri-qiang;LIU Yang;FENG Hai-kuan(School of Surveying and Land Information Engineering,Henan Polytechnic University,Jiaozuo 454000,China;Key Laboratory of Quantitative Remote Sensing in Agriculture of Ministry of Agriculture and Rural Affairs,Information Technology Research Center,Beijing Academy of Agriculture and Forestry Sciences,Beijing 100097,China;National Engineering and Technology Center for Information Agriculture,Nanjing Agricultural University,Nanjing 210095,China)

机构地区：[1]河南理工大学测绘与国土信息工程学院,河南焦作454000 [2]农业农村部农业遥感机理与定量遥感重点实验室,北京市农林科学院信息技术研究中心,北京100097 [3]南京农业大学国家信息农业工程技术中心,江苏南京210095

出　　处：《光谱学与光谱分析》2024年第12期3443-3454,共12页Spectroscopy and Spectral Analysis

基　　金：黑龙江省揭榜挂帅科技攻关项目(2021ZXJ05A05);国家自然科学基金项目(41601346);山西黄河流域矿区采动生态演变驱动机制与协同修复,国家自然科学基金重点项目(U22A20620)资助。

摘　　要：叶面积指数(LAI)是表征作物生长状况的重要指标,因此高效、准确地估算作物LAI可为田间生产管理提供指导。光谱特征能够提供作物反射和吸收波长的信息,而纹理特征则可提供关于作物灰度属性和空间位置关系。以往研究表明仅使用光谱特征估算作物LAI存在一定的局限性,在高LAI水平时,会出现“饱和现象”,使得LAI被低估。为充分发掘无人机多光谱影像信息,对多个波段的纹理信息进行组合,得到多波段组合纹理,探究光谱特征融合多波段组合纹理能否提高LAI估算精度。首先获取马铃薯三个关键生育期的多光谱数据和地面实测LAI数据;然后使用灰度共生矩阵(GLCM)提取各生育期的纹理特征,将多个波段的纹理特征组合;然后分析植被指数、纹理特征和多波段组合纹理与LAI的相关性,并综合相关性和方差膨胀因子优选植被指数;最后融合多波段组合纹理,使用带有参数调优的偏最小二乘回归(PLSR)、岭回归(RR)和K近邻回归(KNR)估算各生育期马铃薯LAI,并与仅使用植被指数的模型进行对比,验证利用多波段组合纹理反演LAI的可行性。结果表明:(1)单波段纹理、双波段组合纹理和三波段组合纹理与LAI的相关性依次增大;(2)马铃薯各生育期优选后的多波段组合纹理与LAI呈现极显著相关,相关系数在0.79~0.83之间;(3)与仅使用植被指数的模型相比,加入多波段组合纹理可以显著提高模型的精度和稳定性。块茎形成期KNR模型的马铃薯LAI估算精度最高,建模R^(2)为0.83,RMSE为0.23 m^(2)·m^(-2),验证R^(2)为0.75,RMSE为0.25 m^(2)·m^(-2);块茎增长期PLSR模型的估算精度最高,建模R^(2)为0.73,RMSE为0.26 m^(2)·m^(-2);验证R^(2)为0.87,RMSE为0.20 m^(2)·m^(-2);淀粉积累期PLSR模型估算精度最高,建模R^(2)为0.73,RMSE为0.31 m^(2)·m^(-2),验证R^(2)为0.84,RMSE为0.25 m^(2)·m^(-2)。该方法可为无人机多光谱组合纹理特征估算马铃薯LAI提�The leaf area index(LAI)is an important indicator for characterizing crop growth,so an efficient and accurate estimation of crop LAI can guide field production management.Spectral features can provide information about the reflected and absorbed wavelengths of crops,while texture features can provide information about the gray-scale attributes and spatial location relationships of crops.Previous studies have shown some limitations in estimating crop LAI using only spectral features,and at high LAI levels,the“saturation phenomenon”occurs,resulting in an underestimation of LAI.To fully explore the information of multispectral images from UAVs,texture information of multiple bands was combined to obtain multiband combined texture and to explore whether the fusion of spectral features with multiband combined texture can improve the accuracy of LAI estimation.Firstly,we obtained multispectral data and ground-truthed LAI data of three key fertility stages of potato;then we extracted the texture features of each fertility stage using the gray-level co-occurrence matrix(GLCM)and combined the texture features of multiple bands;then we analyzed the correlation between the vegetation index,the texture features,and the multi-band combination of textures and LAI,and synthesized the correlations and correlations with LAI,and investigated whether the fusion of spectral information and multiband combination of textures could improve the accuracy of LAI estimation.Then,we analyzed the correlation between vegetation index,texture features,and multiband combined texture and LAI and combined the correlation and variance expansion factors to select the preferred vegetation index;finally,we integrated the multiband combined texture and used partial least squares regression(PLSR),ridge regression(RR)and K-nearest neighbors regression(KNR)with parameter tuning to determine the correlation between the vegetation index and LAI,and then used KNR to estimate the correlation between the vegetation index and LAI.KNR will estimate potato L

关 键 词：无人机 多光谱遥感 多波段组合纹理 叶面积指数 

分 类 号：S25[农业科学—农业机械化工程]