【摘要】 叶面积指数(LAI)是反映不同玉米特性的重要长势指标，可有效辅助玉米新品种的选育。快速、无损和精准地获取玉米LAI,对玉米育种具有重要的指导意义。目前，无人机可见光遥感技术因具有快速、无损和高通量地获取田间作物空间信息的优点，在作物LAI等表型信息获取应用中得到快速发展；然而，其存在的光谱饱和现象以及缺乏光谱参数与LAI之间的响应机理信息，限制了LAI等表型信息估算模型精度的进一步提高；显然，PROSAIL辐射传输模型具有模拟作物理化参数与光谱指标参数之间响应机理的优点，可有效提升作物理化参数反演的潜力。为此，结合无人机数码影像与PROSAIL模型反演夏玉米LAI,以期进一步提高LAI反演模型的精度。以玉米育种试验田的夏玉米为研究对象，利用无人机遥感系统获取拔节期、喇叭口期和抽雄吐丝期的高清数码影像，并结合PROSAIL模型，采用偏最小二乘回归(LSR)、随机森林回归(RFR)和卷积神经网络(CNN)回归方法，构建夏玉米LAI的反演模型。结果表明：(1)基于无人机高清数码影像，采用PLSR回归方法构建的模型精度最优，估算模型和验证模型的R~2、 RMSE和nRMSE分别为0.69, 0.37, 24.28%和0.73, 0.35, 23.26%;(2)基于PROSAIL模型，采用RFR回归方法构建的模型精度最优，估算模型和验证模型的R~2、 RMSE和nRMSE分别为0.98, 0.28, 6.88%和0.87, 0.64, 15.97%;(3)结合无人机高清数码影像与PROSAIL模型，RFR回归方法构建的模型精度最优，估算模型和验证模型的R~2、 RMSE和nRMSE分别为0.98, 0.27, 7.07%和0.87, 0.65, 16.35%,与仅用无人机高清数码影像相比，最优估算模型和验证模型的nRMSE分别降低17.21%和6.91%。研究表明，结合无人机数码影像与PROSAIL模型，有效提高夏玉米LAI反演模型的精度和稳定性，为辅助玉米新品种的选育提供理论指导。更多还原

【Abstract】 The leaf area index(LAI) is an important growth indicator that reflects various maize characteristics and can effectively assist in selecting and breeding new maize varieties. The rapid, non-destructive, and accurate determination of maize LAI is very important in maize breeding. At present, unscrewed aerial vehicle(UAV) visible light remote sensing technology has been rapidly developed in applications for obtaining phenotypic information such as crop LAI because of its advantages in obtaining spatial information about crops in the field in a rapid, non-destructive and high-throughput manner; However, the presence of spectral saturation, due to the lack of information on the response mechanism between spectral parameters and phenotypic information, limits the further improvement of the accuracy of models for estimating phenotypic information; Obviously, the PROSAIL radiative transfer model has the advantage of simulating the response mechanism between crop physicochemical parameters and spectral index parameters, which can effectively enhance the potential of crop physicochemical parameter inversion. Therefore, this study combined UAV digital imagery with the PROSAIL model to invert summer maize LAI to further improve the accuracy of the LAI inversion model. Taking summer maize in the maize breeding experimental field as the research object, a UAV remote sensing system was used to obtain high-resolution digital images at the jointing stage, trumpet stage and tassel emergence stage and combined with the PROSAIL model to construct a summer maize LAI inversion model using partial least squares regression(PLSR), random forest regression(RFR) and convolutional neural network(CNN) regression methods. The results show that(1) based on UAV high-resolution digital images, the model constructed by PLSR regression method has the optimal accuracy, and the R~2, RMSE and nRMSE of the estimation model and validation model are 0.69, 0.37, 24.28% and 0.73, 0.35, 23.26%, respectively;(2) Based on the PROSAIL model, the model constructed using the RFR regression method has the best accuracy, with R~2, RMSE and nRMSE of 0.98, 0.28, 6.88% and 0.87, 0.64, 15.97% for the estimated and validated models, respectively;(3) Combining the UAV high-resolution digital imagery with the PROSAIL model, the RFR regression method constructed the model with optimal accuracy, and the R~2, RMSE and nRMSE of the estimation and validation models were 0.98, 0.27, 7.07% and 0.87, 0.65, 16.35%, respectively. The nRMSE of the optimal estimation model and the validation model were reduced by 17.21% and 6.91%, respectively, compared to using only UAV high-resolution digital imagery. The study shows that combining UAV digital imagery with the PROSAIL model effectively improves the accuracy and stability of the LAI inversion model for summer maize, and provides theoretical guidance to assist in selecting and breeding new maize varieties.更多还原