A Tutorial on GeoAI: Designing Footprint Extraction from NAIP Imagery Using U-Net, Grounding DINO, SAM, and Mask R-CNN

MarkTechPost · 2 months ago · 18

A tutorial presents a complete geospatial AI workflow for extracting building footprints from high-resolution aerial imagery using U-Net semantic segmentation, zero-shot models (Grounding DINO and SAM), and Mask R-CNN instance segmentation. The pipeline processes NAIP imagery through data preparation, model training with ResNet-34 encoder, sliding-window inference, and post-processing steps including orthogonalization and regularization to produce cleaned building polygons. The workflow enables practitioners to train custom footprint extraction models, evaluate performance using IoU and F1 metrics, and compare results across multiple deep learning approaches on real-world geographic areas.