基于Sentinel-1A雷达数据和Sentinel-2A多光谱数据特征融合的地物分类

Ground object classification based on Sentinel-1A radar data and Sentinel-2A multispectral data

基于光学影像受云雨等不良天气的影响,导致在地物分类时易造成数据信息缺失,而雷达影像作为主动式成像,能够较好地克服这一缺陷。笔者选取长春市净月开发区部分地块为研究区域,分别采用最小距离、最大似然和支持向量机3种分类方法,以Sentinel-1A雷达影像和Sentinel-2A多光谱影像为数据源,基于特征融合,提高地物分类精度。结果表明:特征融合后影像的地物分类精度较单一的光学影像有明显提高,且与最小距离和最大似然相比支持向量机分类精度最高。在无云层覆盖的情况下,融合后支持向量机分类精度达到97.94%,较光学影像提高8.11%;在有云层覆盖情况下,融合后支持向量机精度达到77.29%,较光学影像提高12.5%,尤其对水域和建筑区的识别精度有较大提高。

Due to the influence of clouds,rain and other bad weather,optical images are prone to lose data information in ground object classification.Radar images,as active imaging,can well overcome this defect.Part of the Jingyue Development Zone in Changchun is selected as the research area,using three classification methods,namely minimum distance,maximum likelihood and support vector machine,the Sentinel-1A radar image and Sentinel-2A multi-spectral image are taken as the data source to improve the ground object classification accuracy based on characteristics fusion.The results show that the classification accuracy of characteristics fusion images is significantly higher than that of optical images,and the support vector machine accuracy is the highest compared with the minimum distance and maximum likelihood.In the absence of cloud cover,the classification accuracy of support vector machine after fusion reaches 97.94%,which is 8.11%higher than that of optical image.In the case of cloud cover,the accuracy of support vector machine after fusion reaches 77.29%,which is 12.5%higher than the optical image,especially for the identification accuracy of water area and building area.