Heterogeneous Network Based Contrastive Learning Method for PolSAR Land Cover Classification

This paper proposes a Heterogeneous Network based on Contrastive Learning (HCLNet). HCLNet aims to learn high-level representation from unlabeled PolSAR data for few-shot classification according to multi-features. It introduces the heterogeneous network for the first time to utilize different PolSAR features better. Beyond the conventional CL, HCLNet develops two easy-to-use plugins to narrow the domain gap between optics and PolSAR, including beam search and superpixel-based instance discrimination. The pre-trained online network is used for the downstream task by fine-tuning. Experiments demonstrate the superiority of HCLNet on three widely used PolSAR benchmark data sets compared with state-of-the-art methods on few-shot classification. Ablation studies also verify the importance of each component. Besides, this work has implications for how to efficiently utilize the multifeatures of PolSAR data to learn better high-level representation in CL and how to construct networks suitable for PolSAR data better.

Download paper from here

BibTeX formatted citation:

@ARTICLE{10601228,
  author={Cai, Jianfeng and Ma, Yue and Feng, Zhixi and Yang, Shuyuan and Jiao, Licheng},
  journal={IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing}, 
  title={Heterogeneous Network-Based Contrastive Learning Method for PolSAR Land Cover Classification}, 
  year={2024},
  volume={17},
  number={},
  pages={16433-16448},
  keywords={Optical imaging;Optical sensors;Optical scattering;Scattering;Task analysis;Semantics;Remote sensing;Contrastive learning (CL);feature selection;few-shot learning;polarimetric synthetic aperture radar (PolSAR) image classification;superpixel},
  doi={10.1109/JSTARS.2024.3429538}}