基于无人机LiDAR和RGB多特征数据融合的玉米株高估测研究

Research on Maize Plant Height Estimation Based on UAV LiDAR and RGB Multi-feature Data Fusion

摘要:

玉米株高是估算生物量、预测产量的关键因子,也是玉米生长管理的重要参考。本研究选用玉米苗期、拔节期、抽雄期、灌浆期和成熟期的无人机遥感数据,探讨滤波算法及空间分辨率对构建数字高程模型(DEM)的影响;探究LiDAR和RGB 2种数据在玉米株高估测中的优势,通过对2种数据优势的结合,构建一种基于LiDAR、RGB多特征数据融合的回归模型。结果表明,基于改进的渐进三角网加密滤波算法获取的空间分辨率为0.25m的DEM准确性最高,决定系数(R2)为0.9441,均方根误差(RMSE)为0.063m,标准均方根误差(NRMSE)为0.151% ;融合LiDAR和RGB作物高度模型(CHM)第100百分位数、可见光大气抵抗指数(VARI)、红绿蓝植被指数(RGBVI)和冠层体积(CV) 构建的线性回归(LR)模型精度最高,R2为0.9864、RMSE为0.088m、NRMSE为6.331%,相较于最优单特征线性回归模型,R2提升了0.5%,RMSE和NRMSE都降低了16.2%。通过建立融合LiDAR和RGB多特征数据的回归模型,提高了株高估测的精度,该研究为准确估测全生育期的玉米株高提供了一种技术方法。

Abstract:

Maize plant height is a key factor for estimating biomass and predicting yield, and it is also an important reference for maize growth management. UAV remote sensing data of maize seedling stage, jointing stage, tasseling stage, filling stage and mature stage were selected to explore the influence of filtering algorithm and spatial resolution on the construction of digital elevation model (DEM). The advantages of LiDAR and RGB data in maize plant height estimation were explored. By combining the advantages of the two data, a regression model based on LiDAR and RGB multi-feature data fusion was constructed. The DEM with a spatial resolution of 0.25m obtained based on the improved progressive triangulation encryption filtering algorithm had the highest accuracy, with a determination coefficient (R2) of 0.9441, a root mean square error (RMSE) of 0.063m, and a normalized root mean square error (NRMSE) of 0.151%. The linear regression (LR) model constructed by fusing the 100th percentile of LiDAR and RGB crop height model (CHM), visible light atmospheric resistance index (VARI), red, green and blue vegetation index (RGBVI) and canopy volume (CV) had the highest accuracy, R2=0.9864, RMSE=0.088m, NRMSE=6.331%. Compared with the optimal single feature linear regression model, R2 was increased by 0. 5%, RMSE and NRMSE was decreased by 16. 2%. By establishing a regression model that integrated LiDAR and RGB multi-feature data, the accuracy of plant height estimation was improved. The research result can provide a technical method for accurately estimating the plant height of maize in the whole growth period.