This event has ended. Create your own event on Sched.
The 2019 ESIP Winter Meeting has passed. See session descriptions to access meeting content, including presentations, recordings, and key takeaways. See here for info on upcoming meetings.
Back To Schedule
Thursday, January 17 • 11:00am - 12:30pm
Exercising Deep Learning Technique on Earth Datasets for Agriculture

Sign up or log in to save this to your schedule, view media, leave feedback and see who's attending!

Feedback form is now closed.
Session Abstract:
Deep learning (DL) is the hottest method to realize artificial intelligence in many applied science domains. Our ESIPLab project Geoweaver has started to use Deep Learning method in producing crop maps with higher resolution and accuracy than conventional crop map production methods.
The success of DL relies on massive training datasets and powerful compute nodes like Graphics Processing Units (GPU). A good neural network requires careful engineering and considerable domain expertise in network training. It is never easy to fit DL on any Earth dataset. This session will carry out discussion on the research areas, technical details, data sources, and performances of DL in agriculture. We will work on harmonizing and generating a common strategy to connect and prepare Earth datasets for the training/testing of customized deep neural networks to help advance agricultural researches into next level: intelligent agriculture.

Session Takeaways (post meeting):
1) The workflow of deep learning: gather data > choose network type > choose DL library > find powerful hardware > data preprocessing > training > predicting > validation. This workflow can be used in all three aspects of agriculture: monitoring, predicting, and decision making.
2) The quality of crop information in the training datasets is the key to a successful model. The deep learning models must be fed with very accurate information so they can learn those pattern features. Less accurate training datasets will lead to a waste of training time.
3) The trained models have restrictions in time and location. The accuracy is largely related to the quality of training dataset, the chosen deep neural network type, the training process control, and the expertise of the practitioner. Normally, this is a collaboration work which need a stable group to work on it. ESIP machine learning cluster and ESIP Github repos are great platforms and tools for these efforts happen.

avatar for Annie Burgess

Annie Burgess

ESIP Lab Director, ESIP

Thursday January 17, 2019 11:00am - 12:30pm EST
Linden Oak
  Linden Oak, Working Session