From Radar Echoes To Real-Time Recognition: Deep Learning on Another Planet -An Open Source Workflow
summary
Pipeline converts Magellan SAR mosaics into versioned GeoTIFF/GeoJSON layers that geologists can measure and query, with tile prep, speckle-aware augmentation, and human-in-the-loop false‑positive review. Training runs reproducibly in AWS SageMaker containers with fixed seeds and MLflow/DVC logging; models are UNet segmentation + ResNet detector producing pixel masks then vectorized polygons. A single original image is ~60 GB; initial labeling target is 50–100 examples per class and the system already yields usable crater, dome and ridge layers.
AI-written summary · about
chapters (12)
- 0:00Radar mapping on Venus
- 3:58Venus needs radar mapping
- 7:51NASA’s Magellan SAR data
- 12:28Radar echoes to layers
- 15:44Machine learning pipeline
- 18:13Annotation as ML spec
- 20:22Segmentation model strategy
- 22:49Training in cloud lab
- 24:37Visualizing geospatial output
- 26:25Extending beyond Venus
- 28:31Open science future
- 29:14Q&A
related talks
45mFrom Radar Echoes To Real-Time Recognition: Deep Learning on Another Planet an Open Source Workflow
11mTerraKit: Standardising AI-Ready Geospatial Data...
18mDeep learning for Space imagery applications
33mWhat and where? - Machine learning for geospatial image analysis
01h 36mFrom Maps to Models: An End-to-End Journey in Geospatial Machine Learning with Python
26mGeospatial Machine Learning at Scale
44mDemocratizing Climate Science
36mClassification of Satellite Imagery