Go to CVPR 2025 Workshop VLMs4All homepageThe use of multi-modal models and machine learning tech-niques to improve the efficiency and accuracy of geospatial data analysis
Keywords: Large language models, retrieval-augmented generation, computer vision, geospatial analysis, automated imagery processing, contextual data enrichment.
TL;DR: This paper introduces a hybrid geospatial analysis workflow that uses computer vision and generative AI to automate satellite imagery interpretation, improving speed, accuracy, and contextual understanding.
Abstract: Abstract: Geospatial data analysis is heavily reliant on human interpretation of large-scale imagery which leads to constraints in scalability. This study evaluates whether mul-ti-modal models can assist in overhead image understanding by accurately interpreting imagery and automating workflows. A hybrid machine learning solution using Over-sightML (OSML)—an open-source, cloud-based framework—is assessed for its ability to improve geospatial workflows. OSML integrates state-of-the-art computer vision with generative AI capabilities and streamlines preprocessing and detection aggregation. Results indicate that combining domain-specific CV models with foundation models offers a scalable and efficient alternative to manual analysis workflows
Submission Number: 9
Loading