GeoRS-Agent: An Agentic GeoAI Framework for Attribute-Grounded and Explainable Remote Sensing Scene Understanding

Remote sensing (RS) image understanding has significantly advanced through deep learning and vision-language models, yet most approaches still operate mainly as prediction systems that return class labels with limited explanation of the semantic evidence supporting them. Attribute-based representations offer a more interpretable alternative, describing RS scenes in terms of visible objects, dominant colors and geometric patterns. However, these representations are typically used within offline datasets and benchmarking pipelines rather than in interactive systems that enable users to inspect, query and reuse semantic reasoning. This paper presents GeoRS-Agent, an agentic geospatial artificial intelligence (GeoAI) framework for explainable and attribute-grounded RS scene understanding. A vision–language model reports 38 attributes, a deterministic classifier assigns one of 19 classes from class–attribute profiles, and a language model explains the decision citing only the evidence used; an agentic mode exposes these components as tools. On 301 test images, with Qwen2.5-VL 7B throughout, the pipeline reached 58.1% accuracy, against 71.1% with the class list in the prompt, 77.4% with the knowledge base also in the prompt, and 91.7% with reference attributes; with profiles estimated from the model’s own attributes it reached 75.1% in an exploratory analysis; its difference from the knowledge-in-prompt condition was not statistically significant, while the class decision remained inspectable. Explanations were always consistent with the evidence; tools improved knowledge and contrastive queries but not counterfactual ones.

Authors

Institutions

Publication Details

Journal
Computers
Published
2026-11-17
DOI
https://doi.org/10.3390/computers15110846
Primary Topic
Remote-Sensing Image Classification
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

GeoRS-Agent: An Agentic GeoAI Framework for Attribute-Grounded and Explainable Remote Sensing Scene Understanding

Roberto Pierdicca, Marina Paolanti
2 links
Computers
Remote-Sensing Image Classification
article

GeoRS-Agent: An Agentic GeoAI Framework for Attribute-Grounded and Explainable Remote Sensing Scene Understanding

Roberto Pierdicca, Marina Paolanti
article en

Abstract

Remote sensing (RS) image understanding has significantly advanced through deep learning and vision-language models, yet most approaches still operate mainly as prediction systems that return class labels with limited explanation of the semantic evidence supporting them. Attribute-based representations offer a more interpretable alternative, describing RS scenes in terms of visible objects, dominant colors and geometric patterns. However, these representations are typically used within offline datasets and benchmarking pipelines rather than in interactive systems that enable users to inspect, query and reuse semantic reasoning. This paper presents GeoRS-Agent, an agentic geospatial artificial intelligence (GeoAI) framework for explainable and attribute-grounded RS scene understanding. A vision–language model reports 38 attributes, a deterministic classifier assigns one of 19 classes from class–attribute profiles, and a language model explains the decision citing only the evidence used; an agentic mode exposes these components as tools. On 301 test images, with Qwen2.5-VL 7B throughout, the pipeline reached 58.1% accuracy, against 71.1% with the class list in the prompt, 77.4% with the knowledge base also in the prompt, and 91.7% with reference attributes; with profiles estimated from the model’s own attributes it reached 75.1% in an exploratory analysis; its difference from the knowledge-in-prompt condition was not statistically significant, while the class decision remained inspectable. Explanations were always consistent with the evidence; tools improved knowledge and contrastive queries but not counterfactual ones.

ComputersVol. 15(11)
Marche Polytechnic University (IT), University of Macerata (IT)
Remote-Sensing Image Classification
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon