Crop breeding programs have long faced the challenge of accurately collecting phenotypic information. The leaf chlorophyll content is an important growth indicator in rice breeding and is generally measured using a portable chlorophyll meter. In this study, a high-resolution RGB camera and a multispectral camera were mounted on unmanned aerial vehicles (UAVs) to obtain images of 216 hybrid rice varieties. Four different machine learning algorithms and were used to estimate the leaf chlorophyll content in each plot using 16 vegetation indices (VIs) calculated from the UAV-based images. The obtained results demonstrated that the chlorophyll content estimation performance of boosted regression trees (BRT) was better than random forest (RF), support vector machine (SVM), and partial least squares regression (PLSR) models, with R = 0.712 and root mean square error (RMSE) = 1.524. The optimal model was utilized to analyze the rice chlorophyll content in the time series images through the inversion process, revealing a dynamic trend where the level of chlorophyll reached its highest point 82 days after transplantation (DAT). All rice varieties were grouped into four categories (Clusters A, B, C, and D) using nine dynamic indicators extracted from the chlorophyll content trend curve and the K-means clustering algorithm. According to Tukey's HSD tests (p < 0.05), the dynamic indicators showed significant differences between the four categories, especially for the minimum chlorophyll content () and the difference between the maximum and final value (). The relationship between chlorophyll dynamic indicators and grain yield was analyzed and it was found that despite having the highest chlorophyll content and accumulating the most chlorophyll, Cluster A exhibited significantly lower grain yield. This was evident from the largest maximum chlorophyll content () and the difference between maximum and initial value ( - ) values. The result implied that high chlorophyll content during a growing period or at a specific stage does not necessarily result in high yield. The findings in this study can provide new ideas and a basis for hybrid rice breeding.

中文翻译：

---

基于无人机的跨基因型稻叶叶绿素含量动态评估

作物育种计划长期以来一直面临着准确收集表型信息的挑战。叶片叶绿素含量是水稻育种中重要的生长指标，一般采用便携式叶绿素仪进行测量。在这项研究中，高分辨率 RGB 相机和多光谱相机安装在无人机 (UAV) 上，以获得 216 个杂交水稻品种的图像。使用四种不同的机器学习算法，使用根据无人机图像计算出的 16 个植被指数 (VI) 来估计每个地块中的叶片叶绿素含量。结果表明，提升回归树（BRT）的叶绿素含量估计性能优于随机森林（RF）、支持向量机（SVM）和偏最小二乘回归（PLSR）模型，R = 0.712，根均值平方误差 (RMSE) = 1.524。利用优化模型通过反演过程分析时间序列图像中的水稻叶绿素含量，揭示了叶绿素水平在移植后82天（DAT）达到最高点的动态趋势。利用从叶绿素含量趋势曲线中提取的九个动态指标和K均值聚类算法，将所有水稻品种分为四类（聚类A、B、C和D）。根据 Tukey 的 HSD 检验（p < 0.05），四个类别之间的动态指标存在显着差异，特别是叶绿素含量最小值（）以及最大值与最终值之间的差异（）。分析叶绿素动态指标与籽粒产量的关系发现，尽管A簇叶绿素含量最高、叶绿素积累最多，但籽粒产量却显着较低。这从最大叶绿素含量 () 以及最大值和初始值 (-) 值之间的差异可以明显看出。结果表明，在生长期间或特定阶段叶绿素含量高并不一定会导致高产。本研究结果可为杂交水稻育种提供新思路和依据。