search

kevinlacaille / planet_hack_2020_deforestation

Related Project Pitch: https://github.com/planetlabs-community/planet-hack-2020/issues/64
2 stars 4 forks source link

[Proposal] Creating a deforestation database of satellite imagery #20

Open skyprince999 opened 3 years ago

skyprince999 commented 3 years ago

Hi All

I wanted to propose building a dataset of satellite images using the DETERS dataset. As I understand CBERS-4 has a spatial resolution of 20m. ie. each pixel on the image will be a 20sqm. I am planning to use the available data to extract the satellite imagery & then annotate it with the area which has undergone deforestation. These images can then be used as training samples for a segmentation model.

The final goal is to create a data pipeline which will consume daily images from CBERS-4 & identify potential deforestation areas with high accuracy. High-res images can then be used to get a clearer picture.

Would love to know your thoughts? @guy1ziv2 @gilles-morain

gilles-morain commented 3 years ago

@skyprince999 I am not familiar with CBERS-4 data, and currently more focused on increasing the global reusability of the app we started so even though I get the concept, I do not have a clear opinion on your suggestion. From a software architecture perspective, I see it more on the 'notebooks' side of the GitHub project (data preparation and preprocessing) with a potential new column added to the input file of the web application helping users focus on the 'best' rows to check on Planet Explorer

skyprince999 commented 3 years ago

@gilles-morain I can understand that. CBERS-4 is a remote-sensing satellite from which the DETERs dataset (the base data that we were using during the hack.) was built. Some links for reference: https://earth.esa.int/web/guest/missions/3rd-party-missions/potential-missions/cbers https://en.wikipedia.org/wiki/CBERS-4

I got this idea, during one of the zoom calls..... when Guy mentioned that they had to manually work on annotating the data. Building this pipeline will automate and speed-up the discovery of areas of interest .