Multifrequency Polsar Image Classification Using Dual-Band 1D Convolutional Neural Networks

Abstract

In this work, we propose a novel classification approach based on dual-band one-dimensional Convolutional Neural Networks (1D-CNNs) for classification of multifrequency polarimetric SAR (PolSAR) data. The proposed approach can jointly learn from C- and L-band data and improve the single band classification accuracy. To the best of our knowledge, this is the first study that introduces 1D-CNNs to land use/land cover classification domain using PolSAR data. The proposed approach aims to achieve maximum classification accuracy by one-time training over multiple frequency bands with limited labelled data. Moreover, the proposed dual-band 1D-CNN approach yields a superior computational efficiency compared to the deep 2D-CNN based approaches. The performed experiments using AIRSAR PolSAR image over San Diego region at C- and L-bands have shown that the proposed approach is able to simultaneously learn from the C- and L-band SAR data and achieves an elegant classification performance with minimal complexity.