基于随机蕨的极化SAR图像地物分类研究

Research on terrain classification of polarimetric SAR images based on random ferns

传统极化SAR图像地物分类方法通常存在计算效率低和维度灾难等问题,受益于随机蕨分类器的简单性、鲁棒性和处理高维特征空间的能力,文中提出了一种基于随机蕨算法的极化SAR分类框架算法。随机蕨分类器中大量的二元特征捕获了极化SAR图像中地物的空间信息、纹理属性和与其相邻像素的关系。该方法能够在人工标注像素数量较少的情形下对极化SAR图像进行准确、高效的地物分类并且所需要的训练一个随机蕨分类器的时间仅需几十秒。最终的分类实验结果表明,该方法在Oberpfaffenhofen数据集上达到了较好的分类性能和运行效率。

There are problems in traditional Polarimetric SAR(PolSAR)image classification methods,including low computational efficiency and curse of dimensionality problem.Benefiting from the simplicity,robustness and ability of the random fern classifier to handle high-dimensional feature spaces,this article proposes a PolSAR image classification framework based on random ferns algorithm.A large number of binary features in the random fern classifier capture the spatial information,texture attributes and the relationship between the ground objects in the PolSAR image and their neighboring pixels.This method can accurately and efficiently classify PolSAR images with a small number of manually labeled pixels,and the time required to train a random fern classifier is only tens of seconds.The final classification results indicate that this method achieves satisfactory classification performance and operating efficiency in the Oberpfaffenhofen data set.