With the recent release of TensorFlow 2.0, showcased at GEO for Good 2019, there is increased interest in employing an array of neural net approaches to solve various Remote Sensing research questions. A dedicated group from NASA SERVIR has spent the last 5 days exploring the Google Earth Engine, Google Colaboratory, and Google Artifical Intelligence Platform ecosystem applying these modern machine learning techniques to land cover and land use service areas.
The figure showcases the current necessary workflow to deploy a neural net in the cloud. Specifically, there are two main “actions pathways”, displayed in yellow and blue, for leveraging the Colab virtual machine environment or AI Platform respectively. In addition, there are four essential steps necessary for getting a TensorFlow model ready to be used in Google Earth Engine.
More blog posts will follow detailing land cover and land use examples focusing on Machine and Artifical Intelligence approaches and best practices.
Header Photo Credit: Google Earth Outreach
Bade B'kes
Do you have any sample code where Tensor Flow and google earth engine work flow
biplovbhandari
There are plenty on the internet or earth engine forum. You can simply sample your image in earth engine, and use those training samples to train your model in tensorflow.