基于无人机多光谱影像的荒漠草原典型物种识别

Identification of typical species in desert steppe based on unmanned aerial vehicle multispectral images

关键词:  荒漠草原  典型物种  多光谱数据  面向对象影像分析  基于像元影像分析

Key words:  desert grassland  typical species  multispectral data  object oriented image analysis  based on pixel image analysis

摘要:

【目的】 准确获取草原植物物种空间分布信息是草原生态系统生物多样性监测、群落重构与生态功能维持的重要基础。及时准确获取植物物种空间分布可以为草原植物物种信息提取提供有效技术手段。【方法】 文章以无人机多光谱影像为基础，分别在像元尺度和对象尺度上开展了荒漠草原典型物种的信息提取方法研究。像元尺度上先定义样本计算样本可分离性，在选择不同分类器进行分类。而对象尺度上首先进行遥感影像尺度分割研究，选出最佳分割尺度。在此基础上，提取最优特征变量，并采用阈值分类法提取植被信息。【结果】 高分辨率无人机多光谱数据能够为荒漠草原物种信息提取提供有效数据基础。面向对象影像分析技术的表现最好，总体精度85.16%，Kappa系数0.71，其中短花针茅的制图精度和用户精度分别为97.6%和86%；其次是支持向量机机器学习算法，其总体精度80.40%，Kappa系数0.70，短花针茅的制图精度和用户精度分别为90.08%和76.46%；而传统最大似然分类法的识别精度较低，总体精度为74.68%，Kappa系数0.64，短花针茅的制图精度和用户精度分别为72.40和81.96。【结论】 无人机多光谱数据对于集中连片分布的植被物种的识别能力较强，而对零星分布的物种的识别精度并不理想，但该文结果能够为大尺度草原植物物种识别提供一定参考。

Abstract:

[Purpose] To accurately obtain the spatial distribution information of grassland plant species is an important basis for monitoring biodiversity，community reconstruction and ecological function maintenance of grassland ecosystem. To obtain the spatial distribution of plant species timely and accurately can provide effective technical means for the extraction of grassland plant species information.[Method] Based on unmanned aerial vehicle （UAV） multispectral images，information extraction methods of typical species in desert steppe were studied at pixel scale and object scale respectively. At the pixel scale，samples are defined to calculate the separability of samples，and then different classifiers are selected for classification. At the object scale，the segmentation of remote sensing image is firstly studied to select the best segmentation scale. On this basis，the optimal characteristic variables were extracted and the vegetation information was extracted by threshold classification method.[Result] High-resolution UAV multispectral data could provide an effective data base for species information extraction in desert steppe. The object-oriented image analysis technique performed best，with an overall accuracy of 85.16% and a Kappa coefficient of 0.71. The mapping accuracy and user accuracy of Stipa cylindrica were 97.6% and 86%，respectively.The overall accuracy and Kappa coefficient were 80.40% and 0.70，respectively. The mapping accuracy and user accuracy of Stipa cylindrica were 90.08% and 76.46%，respectively. The overall accuracy of the traditional maximum likelihood classification method is 74.68%，and the Kappa coefficient is 0.64. The mapping accuracy and user accuracy of Stipa breviflora are 72.40 and 81.96，respectively.[Conclusion] UAV multi-spectral data has a strong ability to identify concentrated and contiguous vegetation species，while the accuracy of species identification for sporadic distribution is not ideal. However，the results of this study can provide a certain reference for large-scale grassland plant species identification.