CoDEx: Combining Domain Expertise for Spatial Generalization in Satellite Image Analysis

2025-04-28

AI Summary

In satellite image analysis, models suffer significant performance degradation when trained and tested on geographically disparate regions; existing global datasets and methods inadequately address this domain generalization challenge. To tackle this, we propose a Multi-Expert Collaborative Domain Generalization (MECDG) framework: it constructs region-specific expert models for each training domain, explicitly models inter-expert similarity, and enforces prediction consistency regularization—thereby avoiding the performance compromises inherent in monolithic global models. Furthermore, we design a similarity-guided dynamic selection and weighted fusion mechanism, enabling fine-grained, interpretable spatial adaptation. Our work introduces the novel “expert similarity constraint” paradigm. Extensive experiments demonstrate state-of-the-art performance across four major benchmarks—DynamicEarthNet, MUDS, OSCD, and FMoW—outperforming existing domain generalization and adaptation methods. The code is publicly available.

Technology Category

• Machine Learning: Mixture of Experts (MoE)
• Computer Vision: Multi-modal Vision
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Vertical and domain-specific search

Abstract

Global variations in terrain appearance raise a major challenge for satellite image analysis, leading to poor model performance when training on locations that differ from those encountered at test time. This remains true even with recent large global datasets. To address this challenge, we propose a novel domain-generalization framework for satellite images. Instead of trying to learn a single generalizable model, we train one expert model per training domain, while learning experts' similarity and encouraging similar experts to be consistent. A model selection module then identifies the most suitable experts for a given test sample and aggregates their predictions. Experiments on four datasets (DynamicEarthNet, MUDS, OSCD, and FMoW) demonstrate consistent gains over existing domain generalization and adaptation methods. Our code is publicly available at https://github.com/Abhishek19009/CoDEx.

Problem

• Addressing poor model performance in satellite image analysis due to terrain variations
• Proposing a domain-generalization framework with multiple expert models
• Improving accuracy by selecting and aggregating predictions from suitable experts

Innovation

• Train one expert model per training domain
• Learn and encourage consistency among similar experts
• Select and aggregate predictions from suitable experts