Abstract:Supervised deep neural networks (DNNs) have proven to be powerful tools for synthetic aperture radar (SAR) image interpretation tasks. However, they present a formidable challenge in acquiring a substantial amount of labeled data. In this article, we investigate the promising technique of contrastive self-supervised learning for SAR image classification. This approach allows us to take advantage of a large number of available unlabeled images to pretrain an SAR image classification model. Our novel contrastive learning framework conducts both instance-level and cluster-level pretext tasks, which not only enforce consistency between the images and their augmented “views” at the instance level but also their representation within clusters. Besides generating different views through random, low-level image transformations, we proposed two new strategies to construct positive sample pairs to improve contrastive SAR image feature learning: middle-level optical assistance and high-level graph searching. The middle-level optical assistance strategy is inspired by the observation that domain experts typically interpret SAR images with the aid of optical images. This insight spurs us to generate intermediate SAR images as positive samples from geographically matched optical data using CycleGAN. Furthermore, we augment the positive samples of the image with their $k$ -nearest neighbor (KNN) counterparts, following the idea that the KNN samples should belong to the same cluster. Extensive experimental results conducted on the SEN12MS land cover classification benchmark dataset demonstrate that our method is competitive with state-of-the-art self-supervised methods for SAR image classification. Even with only a small amount of labeled data for fine-tuning the model, our method rapidly improves classification performance, surpassing models pretrained on natural image datasets.

Multi-view Subspace Tensor Self-representation for SAR Image Semi-supervised Classification

Deep Co-Attention Network for Multi-View Subspace Learning

Multiview Features Centers Sample Expansion for SAR Image Classification

Self-Supervised Classification of SAR Images with Optical Image Assistance

Two-Stage Multi-Task Representation Learning For Synthetic Aperture Radar (Sar) Target Images Classification

PolSAR Scene Classification Via Low-Rank Tensor-Based Multi-View Subspace Representation.

SAR Image Classification Using Fully Connected Conditional Random Fields Combined with Deep Learning and Superpixel Boundary Constraint

Multiview Attention CNN-LSTM Network for SAR Automatic Target Recognition

Multi-View CNN-LSTM Neural Network for SAR Automatic Target Recognition

Modified Tensor Distance-Based Multiview Spectral Embedding for PolSAR Land Cover Classification

A Semantic Segmentation Method for SAR Image with Assistance of Self-Supervised Scene Classification

Multi-View Tensor Sparse Representation Model for SAR Target Recognition

Self-Supervised Feature Representation for SAR Image Target Classification Using Contrastive Learning

Classification Of Multi-Frequency Polarimetric Sar Images Based On Multi-Linear Subspace Learning Of Tensor Objects

A Novel Semi-Supervised Method Based on Active Sample Proposal and Modified Self-Learning for SAR Target Discrimination

SAR target recognition via sparse representation of multi-view SAR images with correlation analysis

Multi-View Fusion Based on Expectation Maximization for SAR Target Recognition

Self-Supervised Contrastive Learning on Cross-Augmented Samples for SAR Target Recognition

Multi-Block Mixed Sample Semi-Supervised Learning for SAR Target Recognition

A multi-view stable feature extraction network for SAR target recognition

Multi-Aspect SAR Target Recognition Based on Non-Local and Contrastive Learning