Prototype Theory Based Feature Representation for PolSAR Images
- Publisher: Chinese Academy of Sciences
Journal of Radars
(issn: 2095-283X, eissn: 2095-283X)
Technology (General) | Feature representation | Polarimetric Synthetic Aperture Radar (PolSAR) | T1-995 | Prototype theory Unsupervised classification
arxiv: Computer Science::Computer Vision and Pattern Recognition | Computer Science::Graphics
This study presents a new feature representation approach for Polarimetric Synthetic Aperture Radar (PolSAR) image based on prototype theory. First, multiple prototype sets are generated using prototype theory. Then, regularized logistic regression is used to predict similarities between a test sample and each prototype set. Finally, the PolSAR image feature representation is obtained by ensemble projection. Experimental results of an unsupervised classification of PolSAR images show that our method can efficiently represent polarimetric signatures of different land covers and yield satisfactory classification results.