SegDesicNet: Lightweight Semantic Segmentation in Remote Sensing with Geo-Coordinate Embeddings for Domain Adaptation

2025-03-11

AI Summary

To address the poor generalization of semantic segmentation models for remote sensing imagery across geographical regions, weather conditions, and environmental variations—and to overcome limitations of existing methods reliant on expert annotations, specialized hardware, or domain-specific assumptions—this paper proposes a lightweight unsupervised domain adaptation framework. Our core contributions are threefold: (i) we introduce Earth geometry priors by embedding geographic coordinates onto the unit sphere; (ii) we design GRID, a spherical positional encoding that enforces domain invariance via geodesic distance constraints; and (iii) we integrate this prior into a compact CNN backbone to bridge the representational gap between learned features and physical-world geometry. Evaluated on the FLAIR #1 subset, our method achieves a +6.0% mIoU gain with 27% fewer parameters. Further experiments on a custom cross-domain split of the ISPRS Potsdam dataset demonstrate robust generalization across diverse acquisition conditions.

Technology Category

• Computer Vision: Remote Sensing / Geospatial AI
• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

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
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

Semantic segmentation is essential for analyzing highdefinition remote sensing images (HRSIs) because it allows the precise classification of objects and regions at the pixel level. However, remote sensing data present challenges owing to geographical location, weather, and environmental variations, making it difficult for semantic segmentation models to generalize across diverse scenarios. Existing methods are often limited to specific data domains and require expert annotators and specialized equipment for semantic labeling. In this study, we propose a novel unsupervised domain adaptation technique for remote sensing semantic segmentation by utilizing geographical coordinates that are readily accessible in remote sensing setups as metadata in a dataset. To bridge the domain gap, we propose a novel approach that considers the combination of an image's location encoding trait and the spherical nature of Earth's surface. Our proposed SegDesicNet module regresses the GRID positional encoding of the geo coordinates projected over the unit sphere to obtain the domain loss. Our experimental results demonstrate that the proposed SegDesicNet outperforms state of the art domain adaptation methods in remote sensing image segmentation, achieving an improvement of approximately ~6% in the mean intersection over union (MIoU) with a ~ 27% drop in parameter count on benchmarked subsets of the publicly available FLAIR #1 dataset. We also benchmarked our method performance on the custom split of the ISPRS Potsdam dataset. Our algorithm seeks to reduce the modeling disparity between artificial neural networks and human comprehension of the physical world, making the technology more human centric and scalable.

Problem

• Addresses domain adaptation in remote sensing semantic segmentation.
• Utilizes geo-coordinate embeddings to bridge domain gaps.
• Reduces model complexity while improving segmentation accuracy.

Innovation

• Utilizes geo-coordinate embeddings for domain adaptation
• Integrates Earth's spherical nature in image location encoding
• Reduces model parameters while improving segmentation accuracy