TASAM: Terrain-and-Aware Segment Anything Model for Temporal-Scale Remote Sensing Segmentation

2025-09-19

AI Summary

To address SAM’s insufficient zero-shot segmentation generalization on remote sensing imagery—caused by complex terrain, multi-scale objects, and temporal dynamics—this paper proposes TASAM, a lightweight domain-adaptation framework that avoids fine-tuning SAM’s backbone. TASAM introduces three synergistic components: (1) a terrain-aware adapter that incorporates digital elevation model (DEM) priors to enhance geospatial contextual modeling; (2) a temporal prompt generator that leverages change cues to guide segmentation; and (3) a multi-scale feature fusion strategy to improve cross-scale consistency. The framework achieves efficient adaptation with minimal architectural overhead. Evaluated on three benchmark remote sensing datasets—LoveDA, iSAID, and WHU-CD—TASAM significantly outperforms both zero-shot SAM and task-specific supervised models, achieving an average mIoU gain of 8.2% while increasing inference latency by less than 5%.

Technology Category

• Computer Vision: Remote Sensing / Geospatial AI
• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Segment Anything Model (SAM) has demonstrated impressive zero-shot segmentation capabilities across natural image domains, but it struggles to generalize to the unique challenges of remote sensing data, such as complex terrain, multi-scale objects, and temporal dynamics. In this paper, we introduce TASAM, a terrain and temporally-aware extension of SAM designed specifically for high-resolution remote sensing image segmentation. TASAM integrates three lightweight yet effective modules: a terrain-aware adapter that injects elevation priors, a temporal prompt generator that captures land-cover changes over time, and a multi-scale fusion strategy that enhances fine-grained object delineation. Without retraining the SAM backbone, our approach achieves substantial performance gains across three remote sensing benchmarks-LoveDA, iSAID, and WHU-CD-outperforming both zero-shot SAM and task-specific models with minimal computational overhead. Our results highlight the value of domain-adaptive augmentation for foundation models and offer a scalable path toward more robust geospatial segmentation.

Problem

• Adapting SAM for remote sensing terrain challenges
• Addressing multi-scale object segmentation in satellite imagery
• Capturing temporal land-cover changes in geospatial data

Innovation

• Terrain-aware adapter with elevation priors
• Temporal prompt generator for land-cover changes
• Multi-scale fusion strategy for object delineation