摘要
准确反演叶绿素含量对监测水稻长势具有重要意义。由于存在经验模型的缺乏普遍性和物理模型的解不唯一性问题,混合模型成为精确反演叶绿素含量的重要方法,但在此过程中产生冗余的数据集会造成庞大的标记成本。文章基于PROSAIL模型所生成的模拟数据库,将光谱反射率变化率作为特征变量并使用主动学习方法的不同查询策略进行智能采样,通过潜在地抛弃对模型产生负面影响的样本以降低数据集的标记成本,实现水稻冠层叶绿素含量的反演。结果表明,在662 nm到696 nm处的光谱反射率变化率与叶绿素含量相关性系数最高,在反演叶绿素含量时具有更高精度。此外,在主动学习的不同查询策略中,回归池抽样的建模精度最高,R^(2)为0.876,RMSE为1.961。分析表明,主动学习在显著降低标记成本的同时可以提高反演精度,为实现叶绿素含量的精确、快速反演提供依据。
Accurately inverting chlorophyll content is of great significance for monitoring rice growth.Due to the lack of universality in empirical models and the problem of non uniqueness in physical model solutions,hybrid models have become an important method for accurately inverting chlorophyll content.However,hybrid models will generate redundant datasets,resulting in huge labeling costs.This study is based on the simulation database generated by the PROSAIL model,using the rate of change in reflectance between wavelengths‘a’and‘b’as the characteristic variable and using different query strategies of active learning methods for intelligent sampling,by potentially discarding samples that have a negative impact on the model,the labeling cost of the dataset is reduced,and the inversion of rice canopy chlorophyll content is achieved.The results show that the spectral reflectance change rate at 662 nm to 696 nm has the highest correlation with chlorophyll content,and it has better accuracy in inverting chlorophyll.In addition,among different query strategies in active learning,regression pooling sampling has the highest modeling accuracy,R^(2) is 0.876,and RMSE is 1.961.This indicates that active learning can significantly reduce labeling costs while improving inversion accuracy,providing a basis for achieving accurate and rapid inversion of chlorophyll content.
作者
张运
方浩帆
茆红
张露敏
许广涛
ZHANG Yun;FANG Haofan;MAO Hong;ZHANG Lumin;XU Guangtao(School of Geography and Tourism,Anhui Normal University,Wuhu,Anhui 241002,China;Resource Environment and Geography Information Engineering Anhui Engineering Technology Research Center,Wuhu,Anhui 241002,China)
出处
《遥感信息》
北大核心
2025年第2期39-45,共7页
Remote Sensing Information
基金
安徽省重点研究与开发计划(202104g01020004)
安徽省科技重大专项(202003a06020002)
安徽省高等学校科学研究重点项目(2023AH050137)
高分辨率对地观测系统国家科技重大专项(76-Y50G14-0038-22/23)。
关键词
混合模型
光谱反射率变化率
标记成本
智能采样
查询策略
hybrid model
rate of change in reflectance between wavelengths‘a’and‘b’
labeling cost
intelligent sampling
query strategy
