.png){kind=icon}
Return
VEI-Agent: Generating Visual Exploratory Interfaces for remote sensing interpretation
.png){kind=icon}
PRE
AI
DOI：10.1016/j.knosys.2026.116675.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Exploratory interpretation of remote sensing (RS) images is an open-ended process that requires users to flexibly explore image content across multiple perspectives and scales. Although recent large language model-based RS (LLM-RS) methods enable natural language-driven interpretation, they mainly rely on text-only dialogue. This interaction paradigm is inefficient for exploratory interpretation because users must repeatedly reformulate queries to traverse perspectives and scales and integrate fragmented information. To address this limitation, we identify three fundamental patterns of exploratory RS interpretation and propose the Visual Exploratory Interface Agent (VEI-Agent) to support them. Given an RS image and a user instruction, VEI-Agent generates an adaptive, multi-view interface that supports graphical interaction across perspectives and scales. Specifically, it adopts an LLM-centered architecture. Image information is extracted and mapped into the language space, where the LLM jointly reasons over the image information and user instruction to generate textual interpretations and interface design decisions. These decisions are rendered into an interactive interface composed of coordinated texts, tables, and charts with language-driven adaptability. Comprehensive experiments demonstrate that VEI-Agent outperforms existing LLM-RS methods in terms of interpretation quality and interaction efficiency while substantially reducing the cognitive workload of users.
Keywords:
Remote sensing
Exploratory interpretation
Large language model
Visual interactive interface
Journal
K
IF:
7.6
Papers:
1.3W
Citations:
4.5W
Organization
Cited Papers
No cited papers available