Polarimetric SAR Feature Extraction With Neighborhood Preservation-Based Deep Learning

摘要: As an advanced nonlinear technique, deep learning, which is based on neural networks (DNNs), has attracted considerable attentions. In this paper, we propose a novel neighborhood preserved network (NPDNN) for polarimetric synthetic aperture radar feature extraction and classification. The spatial relation between pixels exploited by jointly weighting strategy. Not only the neighbors but also in same superpixel are utilized to weight each pixel. This strategy maintains dependence leading superior homogeneity of terrains without extra computational memory. Moreover, few labeled samples their nearest employed train multilayer NPDNN, preserves local structure reduces number Experimental results synthesized real PolSAR data show that proposed NPDNN can improve classification accuracy compared with state-of-the-art DNNs despite input samples.