






摘要: 為了實現棗樹葉面積指數(LAI)的快速估算,使用無人機多光譜相機獲取新疆阿拉爾墾區棗樹3個生育期的冠層無人機影像,并在地面同步測定樣本點的LAI值,以180種植被指數為基礎構建模型,采用貝葉斯算法中的樹結構Parzen估計器(TPE),提取最優特征組合并優化模型參數,實現模型性能的全面提升,對比分析各模型(CatBoost、RF、DNN、SVR)對棗樹LAI值的監測能力。結果表明,(1)在相同飛行高度下,在坐果期的表現中,4個模型中TPE-CatBoost模型在60 m飛行高度的性能最好,決定系數(R2)為0.867 5,均方誤差(MSE)為0.005 2;(2)利用空間插值法、TPE-CatBoost模型對棗樹LAI進行分析,揭示了整體趨勢和精確的局部分布情況。研究提出的TPE-CatBoost模型實現了墾區棗園棗樹LAI值的有效監測,為墾區棗園的生長監測提供了有效的技術參考。
關鍵詞: 棗樹;葉面積指數;TPE優化算法;CatBoost;特征優選;模型參數選優
中圖分類號: S252+.9 文獻標識碼: A 文章編號: 1000-4440(2024)11-2093-09
Application of multi-altitude UAV multi-spectral imaging in LAI monitoring of jujube trees at different growth stages
HONG Guojun1, ZHANG Ling1, XU Heng2, YU Caili3, HUANG Yufen4,5, FAN Zhenqi4,5
(1.Institute of Regional Development, Jiangxi University of Technology, Nanchang 330200, China;2.Department of Science and Education, Jiangxi University of Technology, Nanchang 330200, China;3.College of Ocean, Shanwei Institute of Technology, Shanwei 516600, China;4.College of Information Engineering, Tarim University, Alaer 843300, China;5.Key Laboratory of Tarim Oasis Agriculture, Ministry of Education, Alaer 843300, China)
Abstract: In order to achieve rapid estimation of leaf area index (LAI) of jujube trees, unmanned aerial vehicle (UAV) multispectral cameras were used to obtain canopy UAV images of jujube trees at three growth stages in Alar Reclamation Area. The LAI values of sample points were measured synchronously on the ground. A model was constructed based on 180 vegetation indices, and the tree structure Parzen estimator (TPE) in Bayesian algorithm was used to extract the optimal feature combination and optimize the model parameters, so as to improve the performance of the model. The monitoring ability of models (CatBoost, RF, DNN, SVR) for jujube tree LAI values was compared and analyzed. The results showed that the TPE-CatBoost model was the best among the four models during the fruit setting period at a flight altitude of 60 meters, with a coefficient of determination (R2 ) of 0.867 5 and a mean square error (MSE) of 0.005 2, respectively. The spatial interpolation method and TPE-CatBoost model were used to analyze the LAI of jujube trees, revealing the overall trend and accurate local distribution. The TPE-CatBoost model proposed in this study can effectively monitor the LAI of jujube trees in reclaimed jujube orchards, providing an effective technical reference for the growth monitoring of jujube in reclaimed areas.
Key words: jujube tree;leaf area index;TPE optimization algorithm;CatBoost;feature optimization;model parameter optimization
棗樹作為新疆地區的重要經濟作物,其健康狀態直接影響棗果的品質和產量,對區域農業經濟發展具有重要意義。作為單位水平地面上單面的葉面積指標,葉面積指數(LAI)是評估作物生長狀況、光合效率、呼吸作用、蒸騰作用等相關生理指標的關鍵參數[1-4]。因此,快速且精確地獲取農作物各生長期的LAI,對病蟲害監測[5]、產量預測[6-9]等田間管理活動至關重要。傳統的實地測量LAI方法不僅具有破壞性,而且通常缺乏實時性和空間分布的準確性[10]。遙感技術在反演LAI方面具有顯著優勢,它不僅能夠長期監測植被的生長狀況,還能實現快速檢測。……