文献类型：会议论文

英文题名：FILL INVALID PIXELS OF THEMATIC MAP USING COMPOSITED IMAGE

作者：Pang Yong[1] Li Zengyuan[1] Fan Shunxiang[1] Meng Shili[1]

第一作者：庞勇

通信作者：Pang, Y[1]

机构：[1]Chinese Acad Forestry, Inst Forest Resource Informat Tech, Beijing 100091, Peoples R China

会议论文集：IEEE International Geoscience and Remote Sensing Symposium (IGARSS)

会议日期：JUL 28-AUG 02, 2019

会议地点：Yokohama, JAPAN

语种：英文

外文关键词：Forest cover map; composited cloud-free image; invalid pixel filling

年份：2019

摘要：Due to the cloud coverage in optical remote sensing data, it is challenging to have a cloud-free image over large region. These inevitable cloud/cloud shadow pixels bring invalid patches in the derived thematic maps, which usually labeled as cloud and cloud shadow types. We use a pixel-level image composition method to generate cloud-free image. For each invalid patch in the thematic map, an enlarged area is extracted in the thematic map and composited image. Then the training data are generated automatically using the thematic map surrounded the invalid patch. The Random Forest (RF) method is used to classify the composited image. The new classification result is used to fill the invalid patch in the thematic map. This method is tested and evaluated in forest cover map in selected sites of the Greater Mekong Subregion (GMS). The result shows it is an efficient way to produce a cloud-free forest cover map for further analysis and applications.