基于半监督学习的SVM-Wishart极化SAR图像分类方法

Semi-supervised Learning for Classification of Polarimetric SAR Images Based on SVM-Wishart

该文针对极化SAR(Synthetic Aperture Radar)图像分类中的小样本问题,提出了一种新的半监督分类算法。考虑到极化SAR数据反映了地物的散射特性,该方法首先利用目标分解方法提取了多种极化散射特征;其次,在协同训练框架下结合SVM分类器构建了协同半监督模型,该模型可以同时利用有标记和无标记样本对极化SAR图像进行分类,从而在小样本时可以获得更好的分类精度;最后,为进一步改善分类结果,在协同训练分类完成后,该方法又利用Wishart分类器对分类结果进行修正。理论分析与实验表明,该算法在只有少量标记样本的情况下优于传统算法。

In this study, we propose a new semi-supervised classification method for Polarimetric SAR（PolSAR）images, aiming at handling the issue that the number of train set is small. First, considering the scattering characters of Pol SAR data, this method extracts multiple scattering features using target decomposition approach.Then, a semi-supervised learning model is established based on a co-training framework and Support Vector Machine（SVM）. Both labeled and unlabeled data are utilized in this model to obtain high classification accuracy.Third, a recovery scheme based on the Wishart classifier is proposed to improve the classification performance.From the experiments conducted in this study, it is evident that the proposed method performs more effectively compared with other traditional methods when the number of train set is small.