结合BRF神经网络的全极化SAR影像分类方法研究

Study on the Whole Polarization SAR Image Classification Based on BRF Neural Network

为了解决模糊支持向量机(FSVM)算法应用于全极化SAR影像分类而产生的聚类中心陷入局部过适应问题,本文提出了一种基于模糊分割理论结合RBF神经网络的全极化SAR影像分类方法。主要利用模糊聚类分割、极化分解、纹理特征提取等,构建待分类地物特征集,并通过SGE进行监督降维,采用降维后的待分类地物极化表征完成RBF分类器训练,实现全极化SAR影像监督分类。最终通过C波段Randsat-2全极化SAR数据进行实测检验,结果表明,该方法使得分类结果区域一致性增强,充分地保存了待分类地物细节信息。

In order to solve the fuzzy support vector machine(FSVM)algorithm in polarization SAR image classification and clustering center into a local adaptation,this paper proposes a combining RBF neural network based on fuzzy segmentation theory full polarization SAR image classification method.Mainly by using fuzzy clustering segmentation,splitting,texture feature extraction,etc.,building classification feature feature set,and supervision by an SGE dimension reduction,the dimension reduction of stay classification feature polarization characterization of complete RBF classifier training,achieve full polarization SAR image classification.The results show that this method has a strong consistency in the classification results,which greatly preserves the details of the classified objects.