智能手机原位牧草生物量估算

Smart-Phone Application in Situ Grassland Biomass Estimation

【目的】生物量是草地生态系统物质和能量基础,是最基本的生态参量。以往基于卫星和航空遥感定量反演草地生物量过于专业化,难以在牧民间推广。因此,本文提出一种用手机近距离拍摄的真彩色图像估算牧草生物量方法,构建牧草生物量估算模型,为牧民方便、快捷、无损地掌握牧场牧草长势提供理论依据和技术支撑。【方法】首先,利用手机超高分辨率真彩色图像,分别基于植被指数、纹理特征以及联合植被指数和纹理特征构建牧草生物量估算特征集合。其次,为防止过多的特征提取带来维度灾难,提出一种XGBoost与序列前向选择相结合的特征选择算法(XGB-SFS),进行特征筛选及最优子集构建。最后,使用随机森林回归和留一法交叉验证对比不同特征集合构建模型的生物量估算效果,分析不同类型特征及XGB-SFS算法在牧草地上生物量(Above Ground Biomass,AGB)估算中的作用。【结果】(1)对比单类型特征构建的模型,基于空间纹理特征的估算模型(R2=0.76)要优于基于光谱植被指数估算模型(R2=0.73),表明纹理特征在超高分辨率牧草AGB估算中具有一定作用;(2)对比特征选择后的模型,联合空谱多类型特征构建模型优于任何一种单类型特征模型(R2=0.83,RMSE=127.57 g·m^(-2),MAE=81.25 g·m^(-2)),表明使用多类型特征构建模型,可一定程度上提高牧草AGB估算精度。(3)对比特征选择前后构建的模型,特征选择后的模型估算AGB效果要明显好于未进行特征选择的模型,且筛选出的特征与牧草生物量之间都存在较高的相关性,表明XGB-SFS能够很好降低数据维度的同时提高牧草AGB估算精度。【结论】手机超高分辨率真彩色图像可以对牧草生物量进行准确估算,本文提出的XGB-SFS算法也能从众多特征中筛选出与牧草生物量相关性较高的特征并提高模型估算精度。与以往专业遥感定量反演草地生物量相比,本文方法具有面向大众、成本低廉、使用方便等优势,研究将手机现场采集的数据与遥感和机器学习方法相结合,可开辟新的视角,支持农业信息化发展。

【Objective】Biomass is the material and energy basis of grassland ecosystem and the most basic ecological parameter.In the past,the quantitative grassland biomass retrieval based on aerospace and aerial remote sensing was too specialized to be popularized among herders.Therefore,this paper proposed a method for estimating grassland biomass by using true color images taken on the phone near the ground,and constructing a grassland biomass estimation model,which provided a theoretical basis and technical support for herders to easily,quickly and non-destructively grasp the growth of grassland in their own pasture.【Method】Firstly,the feature sets of grassland biomass estimation were constructed based on vegetation index,texture features and combined vegetation index and texture features by using the ultra-high resolution true color images of mobile phones.Secondly,in order to prevent dimensional disaster caused by excessive feature extraction,this paper proposed a feature selection algorithm(XGB-SFS)that combined XGBoost and sequence forward selection to perform feature selection and optimal subset construction.Finally,random forest regression and leave-one-out cross-validation were used to compare the biomass estimation effects of different feature sets to build models,and analyze the role of different types of features and XGB-SFS algorithm in grassland AGB estimation.【Result】(1)Compared with the model constructed by single-type features,the estimation model based on spatial texture features(R2=0.76)was better than the estimation model based on spectral vegetation index(R2=0.73),indicating that texture features had a certain role in the ultra-high-resolution grassland AGB estimation;(2)Compared with the model after feature selection,the combined spatial spectrum multi-type feature construction model was superior to any single-type feature model(R2=0.83,RMSE=127.57 g·m^(-2),MAE=81.25 g·m^(-2)),indicating that multi-type feature construction model could improve the accuracy of grassland AGB estimation to a certain extent.(3)Comparing the models building before and after feature selection,the model after feature selection by estimating the AGB effect was significantly better than the model without feature selection,and there was a high correlation between the selected features and grassland biomass,indicating that XGB-SFS could reduce the data dimension and improve the accuracy of grassland AGB estimation.【Conclusion】The ultra-high-resolution true color images of mobile phones could accurately estimate the grassland biomass.The XGB-SFS algorithm proposed in this paper could also select the features with high correlation with the grassland biomass from many features and improve the model estimation accuracy.Compared with the previous professional remote sensing quantitative inversion of grassland biomass,this method had the advantages of facing the public,low cost,and easy to use.The study combined the data collected on the phone with remote sensing and machine learning methods,which could open up new perspectives and support the development of agricultural informatization.