基于无人机点云的玉米冠层三维结构表型对干旱胁迫响应分析

Analysis of Response of Maize Canopy 3D Structure Phenotypes to Drought Stress Based on UAV Point Clouds

摘要:

不同水分梯度下玉米冠层三维结构动态变化规律,评估冠层结构参数对干旱胁迫的区分能力,并解析不同玉米品种冠层结构保持能力差异具有重要作用。本研究以97个玉米育种材料为研究对象,设置正常灌溉（CK）、轻度干旱（D1）和重度干旱（D2）3个水分梯度,利用无人机多视角影像重建冠层三维点云,提取涵盖冠层高度特征、空间覆盖特征和表面结构复杂度的13个冠层结构参数。采用随机森林算法对3种水分处理进行分类识别,并通过KMeans聚类对品种进行冠层结构保持能力分级。研究结果表明,随机森林分类总体准确率为72.5%,冠层高度特征累计重要性达0.481,95百分位高度（0.153）和平均冠层高度（0.114）为区分水分梯度的关键参数。GLCM对比度对干旱胁迫响应最为敏感,CK-D2差异达65%~89%。品种聚类将品种根据旱后表现分为5个类型。冠层三维结构参数对不同水分梯度具有一定区分能力,冠层高度是区分干旱胁迫程度最重要的指标类别,纹理特征可作为冠层高度特征的补充。研究结果为玉米干旱胁迫监测与冠层结构特征筛选提供了技术方法和理论依据。

Abstract:

Aiming to investigate the dynamic changes in the three-dimensional structure of maize canopies under different water stress gradients, evaluate the ability of canopy structural parameters to distinguish drought stress, and analyze differences in canopy structure maintenance among different maize varieties, using 97 maize varieties as study subjects, three water stress gradients were established: normal irrigation (CK), mild drought (D1), and severe drought (D2). RGB imagery captured by an unmanned aerial vehicle (UAV) was used to reconstruct three-dimensional point clouds of the canopy, and 13 canopy structural parameters were extracted, covering canopy height characteristics, spatial coverage characteristics, and surface structural complexity. A random forest algorithm was employed to classify the three water treatments, and K-means clustering was used to categorize varieties based on canopy structure maintenance. The overall accuracy of the random forest classification was 72.5%, with the cumulative importance of canopy height features reaching 0.481. The 95th percentile height (0.153) and average canopy height (0.114) were identified as key parameters for distinguishing between water gradients. GLCM contrast was the most sensitive to drought stress response, with CK-D2 differences ranging from 65% to 89%. Variety clustering classified varieties into five categories. Three-dimensional canopy structure parameters can effectively distinguish different moisture gradients;canopy height was the dominant indicator for drought monitoring, while textural features can provide important supplementary information. The research result can provide technical methods and theoretical basis for monitoring drought stress in maize and screening canopy structure.