【摘要】 [目的]叶面积指数（LAI）是作物生长评估的核心指标，在田间精准管理决策中具有不可替代的关键作用。为了突破传统经验模型的局限，针对水稻不同生育期的冠层结构差异优化模型参数，降低土壤背景、叶片重叠等因素的影响，提高反演精度与效率，实现从遥感数据到叶面积指数的快速转化。[方法]提出基于植被二向性反射统一模型的水稻LAI反演方法。以沈阳农业大学海城精准农业航空科研基地为研究区，采集2023年水稻返青期、分蘖期、拔节期及抽穗期的无人机高光谱数据（400～1 000 nm）与地面实测LAI数据。采用连续投影算法（SPA）筛选特征波段以降低数据冗余。在模型构建方面，通过全局敏感性分析确定模型敏感参数范围，建立多组LAI与冠层反射率的模拟数据集，分别采用查找表法（LUT）与狮群优化算法（LSO）构建反演模型，并与植被指数法、BP神经网络、极限学习机（ELM）及随机森林（RF）等传统方法进行对比验证。[结果]SSPA算法筛选特征波段能有效表征水稻冠层光谱信息；基于植被二向性反射统一模型模拟水稻冠层光谱与实测光谱在400～1 000 nm范围误差较小；基于LSO的LAI反演效果最优，决定系数（R2）达0.779，均方根误差（RMSE）为0.599，显著优于查找表法（R2=0.638,RMSE=0.767）及机器学习方法（BP神经网络R2=0.668,RMSE=0.736;ELM极限学习机R2=0.588,RMSE=0.819;RF随机森林R2=0.649,RMSE=0.756）。[结论]植被二向性反射统一模型凭借明确的物理机制，可有效克服传统数据驱动方法的过拟合等问题，在不同生育期及复杂土壤背景下均保持较高稳定性，为水稻生长动态监测与精准农田管理提供可靠的技术方案，对推动智慧农业的规模化应用具有重要意义。更多还原

【Abstract】 [Objective]Leaf Area Index(LAI) is a core indicator for crop growth assessment and plays an irreplaceable key role in precise field management decision-making. In order to break through the limitations of traditional empirical models, this study optimized model parameters according to the differences in canopy structure of rice at different growth stages, reduced the impacts of factors such as soil background, leaf overlap, improved the accuracy and efficiency of retrieval, to achieve rapid conversion from remote sensing data to leaf area index. [Methods]This paper proposes a retrieval method of rice LAI based on the unified model of vegetation bidirectional reflection. Using the Haicheng Precision Agriculture Aviation Research Base of Shenyang Agricultural University as the research area, unmanned aerial vehicle hyperspectral data(400-1 000 nm) and ground measured LAI data were collected during the seedling regreening, tillering, jointing, and heading stages of rice in 2023. The Successive Projection Algorithm(SPA) was used to screen characteristic bands for reducing data redundancy. In terms of model construction, the range of sensitive parameters of the model was determined through global sensitivity analysis, and multiple simulated datasets of LAI and canopy reflectance were established. The inversion models were constructed using the lookup table(LUT) method and the lion swarm optimization algorithm(LSO), and compared and verified with traditional methods such as vegetation index method, BP neural network,extreme learning machine(ELM), and random forest(RF). [Results]The SPPA algorithm can effectively characterize the spectral information of rice canopy by selecting feature bands; the error between the simulated rice canopy spectrum based on the unified model of vegetation bidirectional reflectance and the measured spectrum is small in the range of 400-1 000 nm; the LAI inversion based on LSO has the best performance, with a coefficient of determination(R2) of 0.779 and a root mean square error(RMSE) of0.599, significantly superior to the lookup table method(R2=0.638, RMSE=0.767) and machine learning method(BP neural network R2=0.668, RMSE=0.736; ELM extreme learning machine R2 =0.588, RMSE=0.819; RF random forest R2 =0.649, RMSE=0.756).[Conclusion]With a clear physical mechanism, the unified model of vegetation bidirectional reflection can effectively overcome the over fitting problems of traditional data driven methods and maintain high stability in different growth periods and under complex soil backgrounds. This study provides a reliable technical solution for dynamic monitoring of rice growth and precise farmland management, which is of great significance for promoting the large-scale application of smart agriculture.更多还原