【摘要】 针对阴影在检测过程中易与水体、深蓝色地物、深灰色道路混淆的问题,对高分二号卫星(GF-2)影像中的典型地物纹理和光谱信息进行统计分析,提出一种基于多尺度分割和随机森林分类的阴影提取算法。通过选取样本数据的阴影指数、亮度、逆差矩、熵、对比度等12个对象特征构建随机森林分类模型,精确提取阴影区域。实验结果显示,本文所提阴影提取算法可以有效地对阴影进行识别,对2个研究区的提取精度分别达到了95.26%和94.35%,可有效解决单一特征下阴影不易与其他地物区分以及阈值分割过程中阈值不易确定的问题。更多还原

【Abstract】 In order to solve the problem of easily confusing with water,dark blue objects and dark grey roads in the process of shadow detection,the texture and spectral information of typical objects in high resolution satellite images(GF-2)were analyzed,and a shadow extraction algorithm based on multi-scale segmentation and random forest classification was proposed.The random forest classification model was constructed by selecting 12 object features of sample data,such as shadow index,brightness,deficit moment,entropy and contrast,to extract the shadow area accurately.The experimental results show that the proposed shadow extraction algorithm can effectively recognize the shadows.And the extraction precision of the two research areas reaches 95.26% and 94.35%,respectively.It can effectively solve the problem that the shadows are not easy to distinguish from other objects under a single feature and the threshold is not easy to determine in the process of threshold segmentation.更多还原