Synthetic Aperture Radar (SAR) Automatic Target Recognition (ATR) systems face significant challenges due to limited labeled data availability and persistent domain gaps between synthetic and measured imagery. This paper presents a comprehensive self-supervised learning (SSL) framework that eliminates dependency on synthetic data while achieving state-of-the-art performance through multi-task pretext training and extensive downstream classifier evaluation. We systematically evaluate our SSL framework across diverse downstream classifiers spanning different computational paradigms and architectural families. Our study encompasses traditional machine learning approaches (SVM, Random Forest, XGBoost, Gradient Boosting), deep convolutional neural networks (ResNet, U-Net, MobileNet, EfficientNet), and a generative adversarial network. We conduct extensive experiments using the SAMPLE dataset with rigorous evaluation protocols. Results demonstrate that SSL significantly improves SAR ATR performance, with SVM achieving 99.63% accuracy, ResNet18 reaching 97.40% accuracy, and Random Forest demonstrating 99.26% accuracy. Our multi-task SSL framework employs nine carefully designed pretext tasks, including geometric invariance, signal robustness, and multi-scale analysis. Cross-validation experiments validate the generalizability and robustness of our findings. Rigorous comparison with SimCLR baseline validates that task-based SSL outperforms contrastive learning for SAR ATR. This work establishes a new paradigm for SAR ATR that leverages inherent radar data structure without synthetic augmentation, providing practical guidelines for deploying SSL-based SAR ATR systems and a foundation for future domain-specific self-supervised learning research in remote sensing applications.
In visual object recognition problems, low light exposure and low-quality images present significant challenges in navigation, surveillance, and image retrieval applications, where reliable feature detection is critical. Although recent deep learning–based image enhancement methods improve visual quality in the pixel domain, these improvements often do not translate to downstream machine vision performance, as important local gradient structures required for stable key point detection are frequently suppressed. In this work, we propose IllumiSIFT, a task-driven dark image enhancement framework that focuses on preserving Scale-Invariant Feature Transform (SIFT) key points by directly learning the Difference-of-Gaussian (DoG) pyramid from low-light image inputs. Unlike conventional pixel-level recovery approaches, the proposed method employs a cascaded residual learning architecture to predict Gaussian-blurred representations at multiple scales, enabling the generation of enhanced DoG images that are inherently aligned with the SIFT detection process. Extensive experiments conducted on the CDVS, Oxford Buildings, and Paris datasets demonstrate that the proposed approach consistently outperforms state-of-the-art enhancement methods in downstream SIFT matching performance under severe low-light conditions. These results confirm that gradient-domain, task-aligned enhancement provides a more effective and practical solution for recognition-centric low-light imaging applications.