多尺度自适应加权与稀疏表示分类相结合的遥感目标识别

发布时间：2018-02-28 02:44

本文关键词： Gabor多尺度自适应加权稀疏表示融合特征　出处：《小型微型计算机系统》2017年09期 　论文类型：期刊论文

【摘要】：针对遥感图像中不同层次的空间结构差异及目标含有不同角度的旋转的情况,提出一种基于Gabor多尺度自适应加权与稀疏表示的遥感目标识别方法.首先对训练样本和待测样本进行Gabor小波变换,对各个方向的Gabor特征进行综合,使它们近似各向同性,根据各尺度特征包含信息量进行自适应加权求和并经过PCA降维求得融合特征,将原始的训练字典改为融合特征字典,从而使字典更加具有判别能力,提高识别率.实验表明,该方法对遥感图像目标识别具有较好的鲁棒性.
[Abstract]:In view of the spatial structure differences at different levels in remote sensing images and the rotation of objects with different angles, A method of remote sensing target recognition based on Gabor multi-scale adaptive weighting and sparse representation is proposed. Firstly, the Gabor wavelet transform is applied to the training samples and the samples to be tested, and the Gabor features in each direction are synthesized to make them nearly isotropic. According to the amount of information contained in each scale, the adaptive weighted summation is carried out and the fusion feature is obtained by reducing the dimension of the PCA. The original training dictionary is changed into the fusion feature dictionary, which makes the dictionary more discriminant and improves the recognition rate. This method is robust to target recognition in remote sensing images.
【作者单位】：长沙理工大学计算机与通信工程学院综合交通运输大数据智能处理湖南省重点实验室;
【基金】：国防"九七三"重点基础研究项目(613XXX0301)资助
【分类号】：TP75

【相似文献】