基于无人机多光谱植被指数与纹理特征的水稻叶绿素含量反演

Accurate Inversion of Rice Chlorophyll Content by Integrating Multispectral and Texture Features Derived from UAV Multispectral Imagery

摘要:

为融合无人机多光谱植被指数和纹理特征实现水稻叶绿素含量估计，本文以大田水稻为研究对象，分别在分蘖期、扬花期及灌浆期等关键生育期进行了无人机多光谱遥感影像和叶绿素含量地面实测值采集；提取了15个多光谱植被指数及35个纹理特征，分析其与水稻叶绿素含量的相关性；并采用仅基于植被指数、仅基于纹理特征和融合光谱及纹理特征等3种建模策略，结合人工神经网络、随机森林、支持向量机及多元线性回归等4种回归建模算法的方式，进行了水稻叶绿素含量精准反演建模分析。结果表明：无人机多光谱植被指数与纹理特征均与水稻叶绿素含量具有显著相关性，其中NGBDI指数与B_M纹理特征相关性最高，皮尔森系数绝对值分别为0.77和0.73；融合无人机多光谱及纹理特征可以有效提升水稻叶绿素含量反演精度，且4种回归算法中人工神经网络的回归估计精度最好，模型验证时调整决定系数为0.72，均方根误差为1.52。融合无人机多光谱及纹理特征可以实现水稻叶片叶绿素含量精准反演，从而为大田水稻精细化管理提供信息支撑。

Abstract:

The emerging unmanned aerial vehicle (UAV) remote sensing technology has gradually become a popular approach to achieve precise management of field crops. Some researches have been conducted on high spatiotemporal resolution, lowcost and accurate monitoring of crop growth. However, there is relatively little research about the estimation of rice leaf green content by integrating UAV multispectral vegetation index and texture features. UAV multispectral remote sensing images and ground measured chlorophyll content of rice were collected during tillering, flowering, and filling growth stages. A total of 50 features, 15 vegetation indices and 35 texture features, were calculated from multispectral images. The max-relevance and min-redundancy (mRMR) algorithm was applied to screen ten vegetation indices and ten texture features from these features. Three modeling strategies were adopted, namely based solely on vegetation indices, based solely on texture features, and based on the combination of vegetation indices and texture features. Four regression modeling algorithms, including artificial neural network (ANN), random forest (RF), support vector machine (SVM), and multiple linear regression (MLR), were used to establish the rice chlorophyll content estimation models. The results showed that both the vegetation indices and texture features were highly correlated with the rice chlorophyll content. Among them, the NGBDI index and the B_M texture feature had the highest correlation, with Pearson coefficients of 0.77 and 0.73, respectively. The fusion of vegetation indices and texture features can effectively improve the estimation accuracy of rice chlorophyll content. Compared with the ANN model based on vegetation indices, the R2 was improved by 0.08 when adding texture features to the models. Among the four regression algorithms, the artificial neural network had the best regression estimation accuracy with R2 of 0.72 and RMSE of 1.52. Therefore, the fusion of vegetation indices and texture features derived from UAV multispectral images can accurately estimate rice chlorophyll content, providing information support for the refined management of rice in the field.