Deep learning for Space imagery applications
summary
Applies deep learning to satellite imagery sized 40,000×400,000 pixels and delivers concrete results: cloud detection cuts error rate by 4× versus prior state of the art, denoising goes from PhD-scale effort to production-quality output in two months, and image-to-information pipelines count aircraft and monitor forests. Uses Docker, Git/GitLab CI, TensorFlow, Apache Spark, Python, GCP machine learning and clusters, running an algorithm farm that trains tens of thousands of configurations with strict versioning of datasets, model structure, loss and optimizer and a deployable continuous learning loop.
AI-written summary · about
chapters (7)
- 0:00Deep learning for space imagery
- 1:49Satellite image deep learning challenges
- 4:16Three key satellite imagery apps
- 7:23Developer insights and algorithm farming
- 10:53Version control for DL models
- 13:41Data exploration and continuous learning
- 16:53Key conclusions and future work
related talks
46mDeep Learning for Satellite Image Processing, Lessons Learned
36mDeep Learning - Segmentation, Identification and Classification
34mDeep Learning at Scale
31mFrom Radar Echoes To Real-Time Recognition: Deep Learning on Another Planet -An Open Source Workflow
40mHow to Implement a Production Ready Deep Learning Model in 12 Weeks
15mKeynote: Incremental Deep Learning For Satellite with KubeEdge...
53mFROM ZERO TO DEEP LEARNING WITH SCALA
45mFrom Radar Echoes To Real-Time Recognition: Deep Learning on Another Planet an Open Source Workflow