基于多源遥感影像融合的武汉市土地利用分类方法研究

准确高效的获取土地利用信息,对于合理利用和开发土地资源具有十分重要的意义。在快速城镇化地区,土地利用活动频繁且密集,土地利用格局演变十分剧烈,增加了城市土地利用精准分类的不确定性;且受环境气候和云雨天气影响增加了有效光学影像获取的难度。为提高城市土地分类精度,该文选取武汉市中心城区为研究案例,以Sentinel-1A和Landsat8 OLI影像为数据源,采用Gram-Schmidt变换方法进行影像融合,选取最大似然、支持向量机、CART决策树、BP神经网络等4种分类方法对融合的影像进行分类,提取了研究区土地利用信息,并对其进行分析。进一步,通过与光学影像的分类结果对比,探究了Sentinel-1A和Landsat8 OLI融合影像在土地利用信息提取方面是否具有优势。研究结果表明:(1)对比其他3种方法,CART决策树分类方法对于融合后的影像分类精度最高,总体分类精度和Kappa系数分别达到88.55%和0.841 4;(2)与光学影像相比,Sentinel-1A和Landsat8 OLI融合影像可以更有效地获取高精度城市土地利用信息;(3)基于多源遥感影像融合的CART决策树分类方法是获取研究区高精度土地利用信息的一种行之有效的技术手段。研究成果可为快速城镇化区域的土地利用分类提供参考。

RESEARCH OF WUHAN CITY LAND USE CLASSIFICATION METHOD BASED ON MULTI-SOURCE REMOTE SENSING IMAGE FUSION

It is of great significance to acquire accurate and efficient land use information for rational use and development of land resources. In areas undergoing rapid urbanization, land use activity is frequent and intensive, and the land use pattern changes very sharply, which increases the uncertainty of the precise classification of urban land use. At the same time, the impact of environment and weather conditions increases the difficulty of obtaining effective optical images. In order to improve the precision of the urban land classification, in this paper we selected the city center area of Wuhan as a case and took Sentinel-1A and Landsat 8 OLI images as the data source, using the Gram-Schmidt transformation method for image fusion. We selected four classification methods to classify the fusion image, including maximum likelihood, support vector machine, CART decision tree and BP neural network to extract the information of land use in the study area. Further, by comparing with the results of the classification of the optical image, we explored whether the Sentinel-1 A and Landsat8 OLI fusion image had the advantage in terms of land use information extraction. The research results showed that:(1) compared with the other 3 methods, the CART decision tree classification method had the highest classification accuracy for the fused image, the overall classification accuracy and Kappa coefficient reached 88.55% and 0.8414; (2) compared with the optical image, Sentinel-1A and Landsat 8 OLI fusion image can obtain high precision urban land use information more effectively; (3) the CART decision tree classification method based on multi source remote sensing image fusion was an effective technique to obtain the high precision land use information in the research area. The results can provide reference for land use classification in the rapid urbanization region.