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.

A Novel Contrastive Learning Model for Aerial Images

Saliency Guided Contrastive Learning on Scene Images

A Novel Self-Learning Network Integrating Contrastive Learning, Perceptual Learning and Masked Image Modelling

A Two-Stage Contrastive Learning Framework for Imbalanced Aerial Scene Recognition

Self-Supervised Classification of SAR Images with Optical Image Assistance

P4Contrast: Contrastive Learning with Pairs of Point-Pixel Pairs for RGB-D Scene Understanding

SAR Image Interpretation Using CNN and Contrastive Learning

Triplet Contrastive Learning Framework With Adversarial Hard-Negative Sample Generation for Multimodal Remote Sensing Images

Enhancing Hyperspectral Image Prediction with Contrastive Learning in Low-Label Regime

Improving Contrastive Learning on Visually Homogeneous Mars Rover Images

Dual-branch Branch Networks Based on Contrastive Learning for Long-Tailed Remote Sensing

Contrastive Learning Based Hybrid Networks for Long-Tailed Image Classification

Unsupervised Representation High-Resolution Remote Sensing Image Scene Classification Via Contrastive Learning Convolutional Neural Network

Improving Contrastive Learning by Visualizing Feature Transformation

Contrastive Self-supervised Representation Learning Using Synthetic Data

Investigating Contrastive Pair Learning's Frontiers in Supervised, Semisupervised, and Self-Supervised Learning

Remote Sensing Images Semantic Segmentation with General Remote Sensing Vision Model via a Self-Supervised Contrastive Learning Method.

Nearest Neighbor-Based Contrastive Learning for Hyperspectral and LiDAR Data Classification

How does Contrastive Learning Organize Images?

Contrastive Attraction and Contrastive Repulsion for Representation Learning

A Novel Contrastive Learning Framework for Self-Supervised Anomaly Detection