基于无人机影像多特征融合的夏玉米LAI动态估计  

Dynamic estimation of summer maize LAI based on multi-feature fusion of UAV imagery

作　　者：余兴娇 樊凯 霍雪飞 殷奇 钱龙 刘政光 张超越 李丽[1,2] 王文娥 胡笑涛[1,2] YU Xingjiao;FAN Kai;HUO Xuefei;YIN Qi;QIAN Long;LIU Zhengguang;ZHANG Chaoyue;LI Li;WANG Wen'e;HU Xiaotao(College of Water Resources and Architectural Engineering,Northwest A&F University,Yangling 712100,China;Key Laboratory of Agricultural Soil and Water Engineering in Arid and Semiarid Areas,Ministry of Education,Northwest A&F University,Yangling 712100,China;School of Engineering,University of Manchester,Manchester,M139PL,UK)

机构地区：[1]西北农林科技大学水利与建筑工程学院,杨凌712100 [2]西北农林科技大学旱区农业水土工程教育部重点实验室,杨凌712100 [3]曼彻斯特大学工程学院,Manchester M139PL3

出　　处：《农业工程学报》2025年第4期124-134,共11页Transactions of the Chinese Society of Agricultural Engineering

基　　金：国家自然科学基金项目(52079113;U2243235);国家重点研发计划项目(2022YFD1900402-01)。

摘　　要：叶面积指数(leaf area index, LAI)是影响作物光合作用并反映作物生长状况的重要参数。及时、准确地监测玉米LAI对提高作物生产力至关重要。该研究旨在探究利用无人机光谱信息、热信息以及冠层形态参数融合以提高玉米多生长阶段LAI估计准确性的潜力,并基于最优估计模型绘制夏玉米LAI反演图,以实现作物水氮精准管理。首先,通过2022—2023年连续2a的田间试验,利用无人机携带的多光谱和热红外传感器采集了不同水氮处理下多个生长阶段玉米冠层图像,并同步测量玉米的生长参数(LAI)。其次,基于冠层光谱、热红外信息和冠层形态参数及其组合建立了LAI估计模型,包括偏最小二乘回归(partial least squares regression, PLS)、反向传播神经网络(backpropagation neural network, BP)和随机森林(random forest, RF)。最后,基于最优估计模型绘制了原位尺度的LAI反演图。研究结果表明:基于冠层光谱信息和热信息的玉米LAI估计与实测LAI的动态变化趋势一致,反映了玉米的生长状况,但单一信息监测多生长阶段玉米LAI具有一定的局限性,估计模型的准确性相对较低。基于光谱信息的LAI估计模型的决定系数(R^(2)),均方根误差(RMSE)分别为0.36~0.61,0.09~0.57;基于温度信息的LAI估计模型的R^(2)=0.25~0.48,RMSE=0.11~0.62。融合多源数据(冠层光谱、热信息和冠层形态参数)提高了玉米LAI的估计精度,3种机器学习模型中,RF模型估计精度最好,其R^(2)=0.814~0.867,RMSE=0.065~0.276。利用RF模型绘制的原位尺度LAI反演图能够准确反映作物的水氮状态。该研究可为无人机平台监测作物生长和水氮管理提供一种可行的方法。Leaf Area Index(LAI)is one of the most crucial influencing parameters on crop photosynthesis during growth status.There is a high demand for the timely and accurate monitoring of maize LAI in the high crop productivity.The purpose of this research was to integrate the Unmanned Aerial Vehicle(UAV)spectral,thermal,and morphological parameters of crop canopy,in order to improve the accuracy of maize LAI estimation at various growth stages.Moreover,an attempt was also made to create the LAI inversion maps for summer maize.The optimal estimation model was achieved to facilitate precise water and nitrogen management.Firstly,the images of maize canopies were collected at multiple growth stages under different water and nitrogen treatments using multispectral and thermal infrared sensors carried by UAVs.The parameters of maize growth(LAI and crop height)were simultaneously measured in field experiments in 2022-2023.Secondly,the LAI estimation models were developed using canopy spectral,thermal infrared data,canopy morphological parameters,and their combinations.Partial Least Squares Regression(PLS),Backpropagation Neural Network(BP),and Random Forest(RF)were also selected using various machine learning algorithms.Finally,the LAI inversion maps at the in-situ scale were created using the optimal RF estimation model.It was found that the maize LAI responded significantly to the water and water-nitrogen treatments,suitable for crop growth monitoring.Among them,the LAI values were ranged from 0.61 to 4.92(0.89 to 5.94)under the N6(N6W2)treatment;The LAI values were ranged from 0.49 to 4.37(0.65 to 4.08)under the N1(N1W0)treatment.Vegetation index(VIs),normalized relative temperature(NRCT),and morphological parameters(CM)of the maize canopy showed stable correlations with the LAI.However,there were some limitations on a single source of information,in order to monitor the LAI of maize at multiple growth stages.The LAI estimation model with the spectral data shared the R^(2) between 0.36 and 0.61,and RMSE between 0.09 and 0.57

关 键 词：无人机 植被指数 归一化相对冠层温度 冠层形态参数 叶面积指数(LAI) 

分 类 号：S126[农业科学—农业基础科学]