基于多源遥感数据融合的矿区土地利用分类方法研究

Study on mining area land use classification method based on multi-source remote sensing data fusion

本文旨在探究利用多源遥感数据融合结合机器学习算法实现土地利用分类的可行性。以冬瓜山矿区及其周边区域作为研究对象,基于NND(Nearest Neighbor Diffusion)融合算法实现哨兵2(Sentinel-2)数据与中巴地球资源卫星04星(CB04)全色数据的融合,并基于融合后的5m分辨率影像实现了建模方法的优选,进行了全区范围内的土地利用分类。实验表明,利用影像融合方法可在保留哨兵2光谱信息的同时有效提升空间分辨率。而在三种机器学习算法下,基于随机森林算法所建分类器整体精度较高,其Kappa系数达到了0.9以上。上述结果则表明,利用多源遥感数据融合与随机森林算法来实现土地利用分类是可行的,该方法不仅可以有效提升影像的空间分辨率,也在一定程度上实现了土地利用分类细化。

The purpose of this paper is to explore the feasibility of land use classification using fused multi-source remote sensing data combined with machine learning algorithm.Taking the Dongguashan mining area and its surrounding area as the research object,based on the NND(Nearest Neighbor Diffusion)fusion algorithm,the Sentinel-2 data and the panchromatic data of the CB04 satellite were fused,and based on the fused 5m resolution image,the optimization of the modeling method was realized,and the land use classification in the whole area was carried out.Experiments show that the image fusion method can effectively improve the spatial resolution while preserving the spectral information of Sentinel 2.Under the three machine learning algorithms,the overall accuracy of the classifier based on the random forest algorithm is high,and its Kappa coefficient reaches more than 0.9.The above results show that it is feasible to use multi-source remote sensing data fusion and random forest algorithm to achieve land use classification.This method can not only effectively improve the spatial resolution of the image,but also achieve the refinement of land use classification to a certain extent.