Urban Tree Canopy Detection Using Satellite and Aerial Imagery

Google Summer of Code 2021

• Author - Aditi Sawant
• Mentor 1 - Nigel Jacob
• Mentor 2 - Michael Lawrence Evans
• Organization - Mayor's Office of New Urban Mechanics (City of Boston)

Abstract

The City of Boston’s Parks Department maintains a comprehensive data set on trees in Boston. However, it’s a manual and laborious process to get the data (such as conducting site visits for tree counts) on a regular basis. We would like to explore the use of satellite and aerial imagery to get a regularly updated census of trees across the city. We would also like to look at the feasibility of determining tree health and the variety of tree species across the city. The ideal outcome would be

• The creation of a new, accurate machine learning model focused on urban tree canopies.
• The creation of a simple web interface for the Parks Department to upload new aerial imagery for analysis
• The ability of the Parks Department to generate a list of statistics on tree counts to ensure that it continues to plants trees in an equitable manner across the city.

Timeline

Community Bonding Period:

17th May - 7th June

• Mentors setup initial video call for primary discussion and explained me about Organization, their current work and my tasks during Community Bonding Period.
• Joined Monum Slack Workspace created by mentors which we will be used as a mode of communication over the GSoC 2021 period.
• Gone through documentation part of DeepForest.
• Successfully installed DeepForest Library locally with Anaconda and tried to run the algorithm on the sample image provided.
• I tried to install DeepForest Tensorflow version on Google Colab but got one error listed here hence installed Pytorch version and it successfully got installed. I successfully run model on sample image provided in Google Colab as well.
• Mentors added me to the new forked branch of DeepForest on Github.

Coding Period:

Research and Development Week 1 to Week 3(7th June - 27th June): *