SARATR-X-v2: Scale-Aware Structural Pre-Training for SAR Foundation Models

2026-07-25

AI Summary

Existing SAR pretraining methods suffer from limited transferability due to reconstruction targets lacking physical stability and multi-scale semantic compatibility. This work proposes a structured pretraining objective that, for the first time, unifies physics-driven speckle invariance with downstream task-oriented multi-scale semantics within a masked image modeling framework. Specifically, six structurally designed extractors with fixed receptive fields—such as blind-spot local aggregation and directional log-ratio regional contrast—generate multi-scale supervision signals, which are then fused via learnable weights to guide reconstruction. The proposed approach achieves state-of-the-art transfer performance across twelve SAR benchmark tasks and reduces representation drift by nearly two orders of magnitude under synthetic speckle perturbations.

Abstract

Masked image modeling has become a dominant paradigm for SAR pre-training, yet the design of the reconstruction target remains fundamentally unsettled. This article argues that a SAR pre-training target should satisfy two conditions to produce transferable representations: (i) physics-grounded stability, i.e., approximate invariance of the target operator to multiplicative speckle inherent in coherent imaging; and (ii) semantic scale compatibility, i.e., coverage of the heterogeneous spatial scales that downstream tasks demand. These two conditions are individually achievable but jointly difficult: physics-grounded stability favors fixed operators, while semantic scale compatibility favors data-driven composition. To this end, SARATR-X-v2 reconciles both within a single design. The target is constructed through fixed structural extractors spanning six receptive fields, from blind-spot local aggregation to directional log-ratio region contrast, and fused via learnable weights into one unified supervision signal for masked reconstruction. On twelve SAR benchmarks across classification, detection, and segmentation, SARATR-X-v2 achieves state-of-the-art transfer performance. Under synthetic speckle variation, the proposed target reduces perturbation drift in the learned representation by nearly two orders of magnitude relative to pixel-space supervision. Taken together, these results establish physics-grounded stability and semantic scale compatibility as a principled framework for pre-training target design under coherent imaging, and suggest that effective SAR pre-training is not about reconstructing more signal, but about reconstructing the right structural target.

Problem

• SAR pre-training
• reconstruction target
• physics-grounded stability
• semantic scale compatibility
• coherent imaging

Innovation

• masked image modeling
• physics-grounded stability
• semantic scale compatibility
• structural pre-training
• SAR foundation models