Tree Species and Condition Identification Challenge

Background

FloCard is a sustainability-focused digital platform that combines digital identity, community engagement, climate action, and nature-based project tools. The platform supports businesses, communities, and sustainability initiatives through features such as digital business cards, sustainability profiles, carbon offsetting, project development, and tools aligned with climate and SDG goals.

FloCard also has a Tree Planters App used by communities, planters, and farms working on afforestation projects.

The app helps users digitally tag and trace trees using:

• geo-tagged single tree images
• Google Map based boundary tagging for clusters of trees or forest areas
• mobile app or mobile browser based field capture
• offline capture support for areas with low or no internet connectivity

Tree tagging usually happens in remote or low-connectivity areas. Users capture the tree image and location first. Later, when internet connectivity is available, the captured assets are synced and edited with details such as species, age, date of plantation, and related information. After review and approval, the tree record becomes a blockchain-based tree asset.

Current Tree Tagging Flow

The tree asset creation process broadly follows three steps:

1. A user captures a tree image and geo-tag in the mobile app or mobile browser, often in offline mode.
2. The asset is synced later and edited with details such as species, age, plantation date, and other metadata.
3. The asset goes through an approval step before becoming a blockchain-based asset.

This process works, but it depends heavily on the quality of the image and the accuracy of manually entered tree details.

Problem Statement

During tree tagging, users may not always know the correct species, approximate age, or visible health condition of the tree.

Incorrect or missing species, age, and health details can reduce asset quality. These fields are also important because they influence carbon or GHG offset estimates. A wrong species or condition can lead to incorrect assumptions about how much carbon a tree may offset in a year.

FloCard is exploring an AI-assisted workflow where tree images captured during the tagging process can help suggest species, visible condition, leaf state, and other reviewable indicators.

Challenge Objective

Build a solution that can analyze a tree image and suggest likely tree species or species group, visible condition, and related metadata.

The solution should support human review rather than treating predictions as final truth.

Expected Outcome

A useful solution should be able to:

• accept tree images or tree crops as input
• suggest a likely species, species group, or supported taxonomy label
• suggest visible health or condition indicators
• identify uncertainty where the model is not confident
• provide reviewable output for users or approvers
• generate metadata for each prediction
• document the approach, assumptions, limitations, and failure cases

Suggested Labels

The solution may include labels such as:

• species or species group
• healthy, stressed, damaged, or uncertain
• young, mature, old, or uncertain
• leafy, leafless, partial foliage, or uncertain
• good image, blurry image, occluded image, low-light image, or uncertain

Contributors may propose a better label structure if they explain the reasoning.

Suggested Approach

Contributors are free to propose their own approach.

Possible approaches may include:

• image classification models
• multimodal vision-language models
• embedding similarity against known species examples
• fine-tuned open-source vision models
• hybrid model plus rule-based review workflows

The solution should explain why the selected approach is suitable, what data it needs, and where human review is still required.

Suggested Stack

Preferred languages:

• Python
• C#

Possible libraries and tools:

• OpenCV
• Pillow
• PyTorch
• TensorFlow
• Hugging Face
• OpenCLIP
• DINOv2
• ONNX Runtime
• ML.NET where applicable

Contributors may use other tools if they explain the reasoning.

Data Access

Images are not included publicly in this repository unless they are open, synthetic, or approved for public use.

Contributors may use:

• open-license tree images
• contributor-created tree images
• approved sample image packs provided by the maintainers
• public tree species references where licensing allows

Contributors who need access to image samples can contact the maintainers by email. Contact [[email protected]]

Do not commit private geo-tagged images, precise location data, private farm or project data, credentials, or restricted-license imagery to the repository.

Submission Guidelines

• Fork the repository.
• Create a feature branch.
• Submit your solution through a pull request.
• Keep the implementation focused on this challenge.
• Do not commit private images, precise location data, credentials, or large generated datasets.
• Include setup and running instructions.
• Include sample output using safe, open, synthetic, or approved images.
• Explain your assumptions and known limitations.
• Mention any additional data you needed or data gaps you found.
• Document how the solution can be customized and scaled.