Skip to content

README.md

Latest commit

 

History

History
42 lines (25 loc) · 1.32 KB

README.md

File metadata and controls

42 lines (25 loc) · 1.32 KB

RTS Segmentation Model v2

Semantic segmentation of Retrogressive Thaw Slumps (RTS) in Arctic satellite imagery for pan-arctic mapping.

Overview

This project trains a deep learning model to detect RTS from PlanetScope basemap imagery (up to 74N) and deploys it for pan-arctic inference to produce an RTS survey map.

Quick Links

Document Description
Data Specification Data sources, labeling rules, split strategy
Training Guide Model architecture, loss, metrics, hyperparameters
Inference Pipeline Deployment workflow, tiling, post-processing
Post-Inference Post-processing, map-making, visualisation, Quality control, failure mode analysis, threshold tuning

Data

  • Training: 2024 PlanetScope Quarterly Basemap (RGB 3m)
  • Inference: 2025 PlanetScope Quarterly Basemap
  • Labels: Refined from ARTS dataset on 2024 imagery(~2–3k positive, ~20–25k negative tiles)
  • Auxiliary (optional): Sentinel-2 NDVI/NIR, ArcticDEM derivatives

Training

Inference

Post-inference

Computation

Dockerization

Todos:

  1. training in multi-scale
  2. explore GEE satellite embedding as input feature
  3. 2025 micro set to test temporal domain shift