Multimodal Data Fusion and Visual Artificial Intelligence for Earth Observation

This special session addresses recent advances in multimodal data fusion and visual artificial intelligence (AI) for Earth observation (EO) and diverse geo-referenced visual data, including weather, climate, satellite, aerial, UAV, and street-view imagery. It focuses on methods and applications for environmental, agricultural, urban, and territorial domains, with particular relevance to sustainability assessment, disaster resilience, and cultural-architectural heritage monitoring.

Visual AI for EO is evolving from single-sensor, task-specific analysis to comprehensive multimodal and multi-task pipelines that combine complementary observables of the built and natural environment. This direction is reinforced by major international initiatives, including ESA Φ-lab’s AI4EO activities and challenges, as well as the ITU-led AI for Good platform and broader United Nations efforts on the use of AI and EO for sustainability, standards, and societal impact.

The proposed session aims to address visual processing challenges that arise when remote sensing and geo-referenced imagery are deployed across these diverse applications. Relevant methodological themes include multimodal fusion, domain adaptation, limited supervision, explainability, uncertainty awareness, robustness, and scalable deployment. The session therefore defines a clear methodological and application-oriented niche, while remaining broad enough to attract contributions from complementary research communities.

The session welcomes contributions spanning methodological advancements and applied research. Topics of interest include, but are not limited to:

The session targets a multidisciplinary audience of researchers and practitioners, bridging methodological innovation and applied sciences. It is specifically designed for technical experts in visual information processing, computer vision, multimodal learning, and geospatial AI, alongside domain specialists in remote sensing, environmental intelligence, urban analytics, digital heritage, and sustainability.

As with EUVIP 2026 regular papers, prospective authors are invited to submit full-length papers of up to 6 pages of technical content, figures, and references through the submission system (page 6 should contain references only). Submitted papers will undergo a double-blind review process, and accepted papers will be presented in a lecture or poster session. Papers presented at the conference will be included in the proceedings and are expected to appear in IEEE Xplore.

The session also welcomes “presentation papers” accepted at major conferences or journals within the past year. To avoid issues of double submission, these papers will not be included in the conference proceedings, but their authors will have the opportunity to present their work to the EUVIP audience. Presentation papers are subject to the standard conference registration.