AI for Trees Challenge
Monitor tree growth on drone and satellite imagery from reforestation projects in Africa. Create an ML model for automated tree detection and segmentation.
Premium participation: Get 1-on-1 Mentoring & GuidanceTry a free mentoring session
Monitoring tree coverage of regreening projects in Africa to increase efficiency of carbon capture
Challenge completed. Watch the recording of the AI for Trees final results presentations to discover that the 3 teams Data Enrichment, Drone Data Modeling and Satellite Data Modeling came up with:
The FruitPunch AI community has teamed up with Justdiggit for the AI for Trees Challenge to track the progress of regreening projects in Tanzania and beyond. Justdiggit and its partners in Tanzania inspire and activate local farmers to start regreening their own land by using a combination of agroforestry and rainwater harvesting techniques. So far 500,000 people were empowered to bring back over 9 million trees on smallholder farmland in the Dodoma and Singida regions of Tanzania.
The progress has to be monitored to measure the success, deploy corrective action where needed and better target next regreening actions to model and verify CO2 capture. To keep track of all these trees, drones and satellites were used to collect data on the ground. The aim is to estimate the tree count and tree cover in project areas on the African continent, based on very high resolution RGB drone footage and multispectral satellite imagery.
Our challenge partner
GOAL: Create a ML model for automated tree detection and segmentation on drone and satellite imagery
We will be developing machine learning models to map individual trees, including very young ones, on fields of smallholder farmers. The subgoals will include:
1. Tree detection based on labeled drone footage
- Improve current models based on the mask-labeled data
- Use it to label more data
2. Tree detection on 2D mapping
- Supervised, by mapping it on rgb data
- Unsupervised, with clustering
3. Use satellite data to detect tree coverage
- Map labeled rgb data on satellite data
- Train models on RGB satellite data
- Use multispectral data to detect trees unsupervised
4. Use casted shadows of trees to calculate their height.
To monitor growth of trees in landscape restoration programs in Africa we’ll apply following technologies / methods:
- Object segmentation
- GIS and RS
- Computer vision
Who are we looking for?
We are looking for data science & AI engineering students and professionals, preferably experienced in computer vision - instance segmentation in particular. Basic GIS knowledge is very welcome.
You can join as a contributor (~12 hours per week commitment for 10 weeks) or coach (2-4 hours per week, only for experienced ML professionals)
We’ll organize 2 masterclasses on relevant topics during the challenge to bring you up to speed.
Did you know
🌿 Increasing the Earth’s forests by an area the size of the United States would cut atmospheric carbon dioxide 25 percent.
🌳 83% of people in sub-Saharan Africa are dependent on land for their livelihoods yet two thirds of the land is highly degraded
🌱 Regreening doesn’t just capture carbon, it prevents soil erosion and improves the livelihood of rural population on the continent.
🌲 In one year, 100 mature trees can remove 53 tons of carbon dioxide and 430 pounds of pollution from the atmosphere
🌴 The average American carbon footprint is 20 tons of CO2 per year – an impact 40 trees can offset.
Application Deadline: 8 July 2022
Challenge Kick-off: 13 July 2022
Midterm Presentation: 8 August 2022
Final Presentations: 19 September 2022