联合无人机数据与Sentinel-2影像的流域制图研究

River Basin Mapping Using Combined UAV Data and Sentinel-2 Imagery

沱江流域是长江上游重点生态屏障建设区,是四川省境内生态资源最丰富的区域,其在地理位置、水文特征、区域气候等方面具有重要生态意义。由于流域范围大、地理情况复杂,使得利用卫星遥感进行流域制图过程中,采集训练样本费时费力,而训练样本的采集在遥感制图监督分类过程中至关重要。基于此,研究一种联合无人机数据与Sentinel-2影像的流域制图方法,将无人机数据作为卫星影像流域制图的训练样本区和验证样本区数据源,以解决实地调查采集样本费时费力的问题。首先将无人机采集的样本区数据,通过影像镶嵌、面向对象的分类、坐标转换和降采样处理流程,得到用于卫星影像制图的训练样本和验证样本;其次将无人机样本的降采样结果作为Sentinel-2卫星影像的训练样本;最后采用面向对象分类得到流域制图结果,研究发现将无人机获得的小区域训练样本用于卫星影像分类,总体精度达92.68%,Kappa系数为0.8668,能够满足大范围流域制图精度要求。

The Tuojiang River Basin is key ecological barrier in the upper reaches of the Yangtze River.It has the most abundant ecological resources in Sichuan province,and important ecological significance in terms of geographical location,hydrological characteristics,and regional climate.Due to its large-scale and complex geographical conditions of river basin,it is time-consuming and laborious to collect training samples,however the collection of training samples is very important in supervised classification of river basin mapping.Therefore,the combination of unmanned aerial vehicle(UAV)data and Sentinel-2 imagery is utilized for river basin mapping,in which UAV data is used as the training and verification sample area of satellite images,in order to solve the problem of time-consuming and labor-intensive samples collection in field surveys.Firstly,UAV data is processed through image mosaicking,object-oriented classification,coordinate transformation and down-sampling to obtain training samples and verification samples.Secondly,UAV down-sampling results are used as the training samples in supervised classification of Sentinel-2 satellite images.Finally,the object-oriented classification is used to obtain large-scale river basin mapping.The experimental results show that training samples obtained by UAV in small areas are used for supervised classification of satellite images,the obtained overall accuracy is 92.68%,and Kappa coefficient is 0.8668,it can meet the accuracy requirements of large-scale river basin mapping.