ArcheoVLM
ArcheoVLM
Automated Discovery of Pre-Columbian Sites in the Brazilian Amazon using GPT-4.1 & GPT-o4 Vision Analysis
Hybrid YOLO + OpenAI GPT-4.1/GPT-o4 Vision-Language Models analyzing Sky-View Factor imagery for archaeological feature detection
LiDAR Sky-View Factor analysis, satellite imagery, SAR data, and historical text integration with GPT-4.1 cluster detection
Cloud-based processing pipeline with GKE Autopilot for large-scale GPT-4.1 vision analysis
Objective
To develop and execute a scalable and technically robust automated pipeline that systematically identifies, verifies, and contextually enriches previously undiscovered pre-Columbian archaeological sites within a large-scale LiDAR dataset of the Brazilian Amazon using GPT-4.1 and GPT-o4 vision analysis.
What Makes This Project Next Level
Unlike traditional archaeological surveys, our approach combines the best quality raw LiDAR data available with Point Transformer V3 for ground classification, custom-tuned YOLOv9 object detection, and dual-function GPT-4o analysis in a fully automated end-to-end pipeline. This methodology overcomes the computational barriers that prevent widespread use of high-resolution LiDAR data and delivers scalable archaeological discovery at unprecedented speed and cost.
Primary Deliverable
An interactive, georeferenced database of potential new archaeological sites. Each entry includes location, confidence scores, detected features, GPT-4.1/GPT-o4 generated descriptions, and comprehensive multi-modal verification packages for expert validation.
Core Technologies
- • Python 3.10+ with comprehensive geospatial libraries
- • Google Cloud Platform with GKE Autopilot
- • ORNL DAAC LiDAR dataset (2008-2018)
- • OpenAI GPT-4.1 & GPT-o4 Vision-Language Models
Key Features
- • Point Transformer V3 for ground classification
- • Sky-View Factor (SVF) visualization processing
- • GPT-4.1 cluster detection and analysis
- • Multi-modal verification with SAR and historical data