相干斑噪声背景下的SAR图像分类方法研究

Classification of SAR Images in the Presence of Speckle

研究了相干斑噪声抑制对合成孔径雷达 (SAR)图像分类的影响。分别采用Kuan自适应滤波和小波变换软门限滤波两种方法进行了相干斑噪声抑制 ;对于SAR图像的分类则采用了图像的灰度以及基于灰度级共生矩阵的 4种纹理特征 ,并利用最大似然分类器进行了监督分类。处理结果表明 ,相干斑噪声的抑制尽管可以提高SAR图像的质量 ,但是由于在相干斑噪声得到抑制的同时 ,地物的固有结构信息也受到损失 ,因此分类精度提高甚微 ,在某些情况下甚至有所下降。针对这种情况 ,提出了一种改进的特征提取方法 ,将基于原图像的灰度级共生矩阵提取的纹理特征与滤波后图像的灰度特征进行组合用于分类。实验结果表明 ,改进的特征提取方法提高了SAR图像的分类精度。

In this paper, we investigate the effect of speckle reduction on the classification of SAR images. The adaptive Kuan filter and wavelet soft-thresholding filter are respectively used in speckle reduction. The feature vector is composed of tone of image and four texture features based on the gray-level co-occurrence matrix (GLCM). The maximum likelihood classifier is used in image classification. The classification accuracy of the filtered images is compared with that of the unfiltered images. The results show that although the quality of the image improves, the classification accuracy increases slightly after speckle reduction, and even decreases in some cases. This is due to the loss of some structural information in the course of filtering. Accordingly, we propose an improved feature extraction scheme, adopting the tone of filtered image combined with the texture features based on the GLCM of unfiltered image to form the feature vector. The experimental results show that the improved scheme can enhance the performance of classification.