Stacked Sparse Autoencoder in PolSAR Data Classification Using Local Spatial Information

• Published in: IEEE geoscience and remote sensing letters Vol. 13; no. 9; pp. 1359 - 1363
• Main Authors: Zhang, Lu, Ma, Wenping, Zhang, Dan
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01-09-2016; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Classification; Cost function; Data mining; Deep learning; Feature extraction; Image classification; Image processing; Learning; local spatial information; Machine learning; Pixels; polarimetric synthetic aperture radar (PolSAR); Remote sensing; Scattering; sparse; Speckle; stacked sparse autoencoder (SSAE); Synthetic aperture radar; Training
• ISSN: 1545-598X; 1558-0571
• DOI: 10.1109/LGRS.2016.2586109

Terrain classification is an important topic in polarimetric synthetic aperture radar (PolSAR) image processing. Among various classification techniques, the stacked sparse autoencoder (SSAE) is a kind of deep learning method that can automatically learn useful features layer by layer in an unsupervised manner. However, the scattering measurements of individual pixels in PolSAR images are affected by the speckle; hence, the performance of pixel-based classification approaches would be poor. In this situation, a novel framework is proposed to learn robust features of PolSAR data. The local spatial information is introduced into SSAE to learn the deep spatial sparse features automatically for the first time. Furthermore, the influences of the neighbor pixels on the central pixel are controlled depending on the spatial distances from the neighbor pixels to the central pixel. Experimental results with fully PolSAR data indicate that the proposed method provides a competitive solution.