Sun Ho Ro
← All projects
Computer VisionRemote SensingHydrographyGeospatial Production

AI Culvert Detection

A deep learning object detection workflow designed to locate and attribute culvert features for the USGS 3D Hydrography Program.

Client
United States Geological Survey
Location
Germantown, MD
Year
2023-2024
Role
Geospatial Specialist

Context

The USGS 3D Hydrography Program supports modernization of national hydrography data. Culverts are important features because they influence hydrologic connectivity, but they can be difficult to identify consistently across elevation and imagery datasets.

Challenge

Culverts are small, visually variable, and often embedded in complex terrain or infrastructure settings. A useful detection workflow needed to work within broader geospatial production pipelines and support attribution for derivative products within the 3DEP Quality Level 2 program.

My Role

I led development support for an AI-based culvert detection workflow. My work focused on streamlining object detection methods and integrating AI-assisted identification into the broader 3DHP production process.

Approach

The workflow used deep learning object detection to identify culvert features from elevation and imagery-derived datasets. It was designed to support efficient location and attribution of culvert features as inputs to hydrography production.

Output

The project produced an AI-assisted culvert detection workflow aligned with 3DHP production needs.

Impact

The workflow improved the efficiency of culvert identification and supported modernization of hydrography data production. It showed how targeted computer vision can support national-scale geospatial datasets when paired with production-aware QA/QC.

Tech & Techniques

Deep LearningObject DetectionElevation DataImagery3DHP3DEP QL2