用高分一号卫星数据识别多时相山区积雪

Snow Cover Extraction from Multi-temporal GF-1 Satellite Imagery in Mountain Areas

针对目前从高分辨率遥感图像中提取多时相山区积雪信息精度较低、效率不高的问题,基于高分一号卫星数据,提出了一种基于多视图的多时相山区积雪提取方法。该方法基于GF-1卫星3个时相的图像数据,将多时相遥感图像视为多个视图,通过空间约束构建积雪识别多视图。针对山区阴影的巨大影响的问题,提出了将积雪分为非阴影区积雪和阴影区积雪2个类别,分别进行训练样本的选取。仅通过一次样本选取,运用旋转森林算法,基于多视图构建面向多时相的识别模型,实现多时相遥感图像的积雪识别。实验结果表明,3个时相识别结果的F值分别达到0.941、0.951和0.945,精度较高,且具有较高的效率,具有实际应用价值。

It is difficult to extract snow cover from multi-temporal high spatial resolution remotely sensed imagery, where the precision and efficiency is relatively low. This study provides a method which can extract multi-temporal snow cover from GF-1 based on multi-view model. Based on three GF 1 satellite images, single image was treated as a view, and multi views were built for multi-temporal snow recognition through extracting the unchanged part of multi-temporal images. In order to overcome the severe influence of terrain shadows,labeled samples of snow in sunlight and snow in shadow were selected separately. Only by selecting labeled samples once, three classifiers were built on different views based on rotation forest algorithm. Then, three classifiers were applied to the classification of three multi temporal images. The accuracy verification showed that the F score of three images are 0. 941,0. 951,and 0.945,respectively. In addition, the efficiency is relatively high.