摘要
合成孔径雷达(SAR)图像具有丰富的纹理信息,这些纹理信息能反映地物空间结构关系。当前纹理特征被广泛应用于SAR图像分类和SAR图像分割中。受成像因素影响,直接采用从SAR图像中提取的纹理特征效果不够好。为避免传统先滤波再提取纹理特征的方法对纹理、边缘信息造成损失,提出了一种先提取SAR图像纹理特征,再利用Robust PCA方法对纹理特征去噪的新方法,最后采用Kmeans聚类方法检验RPCA处理后的纹理特征表达效果。实验结果表明该方法能将聚类正确率从82%提高到84%。
SAR（synthetic aperture radar） images contain abundant texture information, which reflects the spatial structure of ground objects. Texture features have been widely used in SAR image classification and segmentation. However, due to effects of the imaging process, classification and segmentation results are not satisfied using texture features extracted directly from SAR image. To avoid the texture information loss using filters before texture features extraction, a new method is proposed that first extracting texture features; then applying the Robust PCA to denoising texture features. Finally Kmeans is used to testify the denoised texture features. Experiment results show that the accuracy rate is improved from 82% to 84%.
出处
《电子技术（上海）》
2015年第4期6-9,5,共5页
Electronic Technology
基金
国家自然科学基金重点项目"高分辨率SAR测试库及数据质量评估"支持(61331015)
依托单位为"上海交通大学"
本人所在实验室承担了该项目的子课题
