鄱阳湖南矶湿地景观信息高分辨率遥感提取

鄱阳湖南矶湿地是亚热带典型过水性湿地,由于该区域水文情况复杂,且泥滩、沼泽和疫水（血吸虫）分布较广,导致野外考察验证工作困难,使用传统的遥感信息提取方法很难保证该地区湿地景观的提取精度。本文以高分一号影像为数据源,综合运用数字高程模型（DEM）、归一化植被指数（NDVI）、归一化水体指数（NDWI）等辅助数据,采用面向对象分类方法,对鄱阳湖南矶湿地景观信息进行提取研究,并取得了较好的分类效果。研究结果表明：（1）基于国产高分辨率影像的面向对象分类,既兼顾了国产高分辨率影像光谱、空间、结构、纹理信息,又综合利用多源辅助数据参与到分类计算中,分类精度得到明显的提升;（2）基于面向对象与多源数据分类方法对湿地混合像元有较好地识别能力,可获得较高的总体分类精度（94.3275%）和Kappa系数（0.9324）,说明利用多源数据的面向对象方法提取湿地信息是可行的,其分类结果具有较高的准确性和可信度,较好地解决了过水性湿地景观分类问题;（3）该分类方法弥补了单一遥感影像分类方法的不足,对研究国产高分卫星在提取过水性湿地景观信息方面具有重要的参考和实际意义。最后,分析了多源数据面向对象分类尚待解决的问题和下一步的研究方向。

The Wetland Information Extraction Research of Nanji Wetland in Poyang Lake Based on High Resolution Remote Sensing Image

The Nanji wetland of Poyang Lake is a typical wetland of the water-carrying type. It is quite difficult to accurately extract the information of wetland landscape from remote sensing images with the traditional information extraction approaches due to the complicated hydrology conditions and arduous field verification, and moreover, the mud flat, swamp and infested water (schistosomiasis) is widely distributed in the region. This study chose GF image as the data source, with the auxiliary data of digital elevation model (DEM), normalized difference vegetation index (NDVI) and normalized difference water index (NDWI) to ensure the accuracy of information extraction. And it uses the object-oriented classification method to extract the landscape information of the Nanji wetland, which achieved some reasonable classification results indicating in the following part; (1) Based on the object-oriented classification of domestic GF satellite remote sensing images, which contains the spectral, spatial structure and texture features, this study makes a comprehensive utilization of the multi-source data in the classification calculation, and the precision of its classification result is significantly higher than the object-oriented classification of single-source; (2) The object-oriented method of multi-source remote sensing data has better distinguish ability for mixed pixels. It obtains a higher overall accuracy of 94.3275% and a Kappa coefficient of 0.9324, which indicates a distinctive high degree of accuracy and reliability. It has effectively solved the classification problem in extracting the wetland landscape of water-carrying; (3) This method makes up the deficiency in the object-oriented classification method of single-source remote sensing image, and it acts as an important reference and has the practical significance for effectively extracting the information of water-carrying wetland landscape through the domestic GF remote sensing image. Finally, we put forward some issues to be resolved and illustrated the future research direction of the object-oriented classification of multi-source data.