New AI framework WILD-SAM adapts the Segment Anything Model to detect landslides in radar imagery by overcoming phase ambiguity challenges.

WILD-SAM introduces a Phase-Aware Mixture-of-Experts (PA-MoE) Adapter to align spectral distributions in wrapped InSAR interferograms

The framework uses Wavelet-Guided Subband Enhancement (WGSE) strategy to preserve high-frequency fringes critical for detecting landslide boundaries

The solution addresses the spectral domain shift problem that prevented the Segment Anything Model (SAM) from working directly on wrapped phase radar data

Parameter-efficient fine-tuning approach enables efficient geohazard monitoring by detecting slow-moving landslides automatically from radar interferograms