基于谱直方图的遥感图像分层次多尺度植被分割

发布时间：2018-05-12 12:37

本文选题：遥感图像 + 植被分割　；参考：《国土资源遥感》2017年02期

【摘要】：在遥感图像分割中,植被是重要的一类对象,植被细分割一般有3个目标,按尺度分为乔木、灌木和草与苔藓。针对单一层次多分类方法不能充分利用植被目标不同纹理尺度实现精确的多分类问题,提出了一种基于谱直方图的遥感图像分层次、多尺度植被分割方法。首先用归一化植被指数(normalized difference vegetation index,NDVI)提取出遥感图像中的植被区域,然后再对该区域分层实现多个二分类算法、完成多分类操作。在每个分类层次,利用目标的先验知识和纹理尺度选择纹理滤波参数,对滤波结果提取各子块图像的谱直方图用以表达纹理特征,从而实现1个层次的分割。实验结果表明,该方法较好地利用了植被各层次目标的先验知识和纹理尺度,使得对纹理滤波器的增强处理更具针对性;谱直方图的特征区分度更大,使得植被细分割精度明显提高。
[Abstract]:In the segmentation of remote sensing images, vegetation is an important kind of object. The fine segmentation of vegetation has 3 targets, which are divided into trees, shrubs and grass and moss according to the scale. In view of the single hierarchical and multi classification method, the precise multi classification problem can not be realized by the different texture scales of the vegetation target. A spectral histogram based remote sensing image is proposed. Hierarchical, multi-scale vegetation segmentation method. Firstly, the vegetation areas in remote sensing images are extracted with normalized difference vegetation index (NDVI), and then multiple two classification algorithms are implemented for the stratification of the region, and multiple classification operations are completed. Texture filtering parameters are used to extract the spectral histograms of each sub block image to express texture features and achieve 1 levels of segmentation. The experimental results show that the method makes good use of the prior knowledge and texture scale of various vegetation levels, making the enhancement processing of the texture filter more pertinent; the spectral histogram is special. The larger the index area, the better the precision of vegetation segmentation.

【作者单位】：辽宁师范大学计算机与信息技术学院;
【基金】：国家自然科学基金项目“适应空间信道传输的星载高光谱海岸带影像压缩编码研究”(编号:41271422) 辽宁省教育厅自然科学基金项目“遥感图像多尺度植被分割技术研究”(编号:L2012379)共同资助
【分类号】：TP751

【相似文献】