基于高光谱特征参数的马铃薯块茎形成期叶片含水量定量监测模型  

Quantitative Monitoring Models of Potato Leaf Water Content in Tuber Formation Period Based on Hyperspectral Characteristic Parameters

作　　者：苏亚拉其其格 张振鑫 李卓凌 樊明寿[2] 贾立国[2] 赵金花 Suyala Qiqige;ZHANG Zhen-xin;LI Zhuo-ling;FAN Ming-shou;JIA Li-guo;ZHAO Jin-hua(Department of Botany,College of Grassland and Resource Environment,Inner Mongolia Agricultural University,Huhhot 010019,China;Department of Plant Physiology,College of Agronomy,Inner Mongolia Agricultural University,Huhhot 010019,China)

机构地区：[1]内蒙古农业大学草原与资源环境学院植物学系,内蒙古呼和浩特010019 [2]内蒙古农业大学农学院植物生理系,内蒙古呼和浩特010019

出　　处：《光谱学与光谱分析》2025年第3期774-783,共10页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(31960388);内蒙古自然科学基金项目(2023LHMS03046,2018BS03002);内蒙古农业大学优博引进项目(NDYB2017-19)资助。

摘　　要：合理水分供应是马铃薯获得高产及优质块茎的必要前提。为实现马铃薯需水关键期的快速水分诊断,采用高光谱遥感和机器学习研究实时监测马铃薯块茎形成期植株水分状况的方法,为干旱地区马铃薯水分高效管理奠定基础。为了构建监测精度高、更具普适性的马铃薯块茎形成期叶片含水量定量估算模型,测定了马铃薯冠层高光谱反射率及叶片含水量数据,筛选响应马铃薯叶片含水量的特征光谱参数,建立预测叶片含水量的偏最小二乘回归、支持向量机和BP神经网络模型。结果表明:针对马铃薯叶片含水量监测,筛选725、856和1000 nm等13个敏感波段,521、555和570 nm等11个特征光谱一阶导数和MSI、NDII、PSRI等7个特征光谱指数。构建的三种模型均能精确定量马铃薯块茎形成期的叶片含水量,说明上述组合光谱特征参数具有较强的实用型。采用全生育期叶片含水量和高光谱数据选择的特征光谱参数对马铃薯关键生育期叶片含水量的定量监测普适性更高,其中BP神经网络模型预测精度最佳。研究结果可以实时、准确地监测马铃薯叶片含水量,对于马铃薯植株水分状态的评估具有重要价值,为马铃薯的快速水分诊断及节水灌溉推荐提供技术支撑。A reasonable water supply is a prerequisite for potatoes to achieve high yield and high-quality tubers.To realize the rapid water diagnosis during the critical period of potato water demand,we used hyperspectral remote sensing and machine learning to study the real-time monitoring of plant water status during the potato tuber formation period to lay the foundation for efficient water management of potatoes in arid areas.This article aims to construct a quantitative estimation model for leaf water content during potato tuber formation with high monitoring accuracy and greater universality.The data of canopy hyperspectral reflectance and leaf water content were measured,and the characteristic spectral parameters that respond to the moisture content of potato leaf finally constructed the Partial least squares regression,Support vector machine,and BP neural network models of leaf water content based on the hyperspectral characteristic parameters.The results showed that:For the monitoring of potato leaf water content,screened the 13 sensitive bands such as 725,856,1000 nm,etc;11 characteristic spectral first-order derivatives such as 521,555,570 nm,etc.;and 7 characteristic spectral indices such as MSI,NDII,PSRI,etc.The three established models that are based on the above characteristic spectral parameters can accurately quantify the leaf water content of potatoes in the tuber formation stage,which means these combined spectral characteristic parameters have strong practicality;Moreover,the use of characteristic spectral parameters screened from full growth stage leaf moisture content and hyperspectral data had higher universality in quantitative monitoring of leaf water content in potato during the critical growth stages.The BP neural network model had the highest prediction accuracy in monitoring leaf moisture content during the tuber formation period.Therefore,this study s results can monitor potato leaves water content in real-time and accurately,which was of great value for evaluating the water status of potato p

关 键 词：高光谱 特征参数 马铃薯 叶片含水量 定量监测 

分 类 号：S365[农业科学—作物栽培与耕作技术]