WILD-SAM: Phase-Aware Expert Adaptation of SAM for Landslide Detection in Wrapped InSAR Interferograms

2026-04-16 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision, Environmental Science & Earth Systems · Depth: Expert, quick

Summary

WILD-SAM is a new parameter-efficient fine-tuning framework designed for high-precision landslide detection directly from wrapped Interferometric Synthetic Aperture Radar (InSAR) interferograms. This model addresses the challenges of severe phase ambiguity and complex coherence noise that hinder direct application of models like Segment Anything Model (SAM) to wrapped phase data. WILD-SAM integrates a Phase-Aware Mixture-of-Experts (PA-MoE) Adapter into the frozen encoder to align spectral distributions, and a Wavelet-Guided Subband Enhancement (WGSE) strategy to generate frequency-aware dense prompts. The PA-MoE Adapter uses dynamic routing across convolutional experts to aggregate multi-scale spectral-textural priors, while WGSE employs discrete wavelet transforms to disentangle high-frequency subbands and refine directional phase textures. Evaluated on the ISSLIDE and ISSLIDE+ benchmarks, WILD-SAM achieved state-of-the-art performance in target completeness and contour fidelity.

Key takeaway

For geohazard monitoring teams utilizing InSAR data, WILD-SAM offers a significant advancement in automated landslide detection. Its specialized architecture, which overcomes phase ambiguity and noise, means you can achieve higher precision in identifying slow-moving landslides. Consider integrating WILD-SAM into your geohazard monitoring workflows to improve the accuracy and efficiency of early detection and risk assessment.

Key insights

WILD-SAM adapts SAM for landslide detection in InSAR data by integrating phase-aware spectral alignment and wavelet-guided prompt enhancement.

Principles

• Address spectral domain shifts in transfer learning.
• Exploit multi-scale spectral-textural priors.
• Refine directional phase textures for boundary integrity.

Method

WILD-SAM integrates a PA-MoE Adapter into SAM's frozen encoder for spectral alignment and uses a WGSE strategy to generate frequency-aware dense prompts via discrete wavelet transforms.

In practice

• Apply PA-MoE for domain adaptation.
• Use wavelet transforms for frequency-aware prompting.

Topics

• WILD-SAM
• Landslide Detection
• Wrapped InSAR Interferograms
• Segment Anything Model
• Phase-Aware Mixture-of-Experts

Best for: AI Scientist, Computer Vision Engineer, Research Scientist

Related on AIssential

• When Less Is More: Simplicity Beats Complexity for Physics-Constrained InSAR Phase Unwrapping — Simpler convolutional architectures outperform complex attention-based models for physics-constrained geophysical regression tasks.
• LandslideAgent with Multimodal LandslideBench: A Domain-Rule-Augmented Agent for Autonomous Landslide Identification and Analysis — Domain-rule-augmented agents enhance VLM performance for complex geological hazard analysis by integrating specialized datasets and control mechanisms.
• Domain-Guided Prompting of the Segment Anything Model for Seismic Interpretation: The Role of Attributes, Visualization, and Hybrid Prompts — Zero-shot SAM adaptation for seismic interpretation is achieved by combining domain-guided visualization with hybrid prompting.
• deFOREST: Fusing Optical and Radar satellite data for Enhanced Sensing of Tree-loss — Fusing optical and SAR satellite data with a novel anomaly detection method enhances deforestation monitoring, especially in cloudy areas.
• Implicit spatial-frequency fusion of hyperspectral and lidar data via kolmogorov-arnold networks — KANs enable adaptive, non-linear fusion of HSI and LiDAR data across spatial and frequency domains.
• OmniCD: A Foundational Framework for Remote Sensing Image Change Detection Guided by Multimodal Semantics — OmniCD unifies remote sensing change detection with multimodal semantic guidance, achieving robust, general-purpose performance across diverse scenarios.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.