Savanna Tree AI

Access the data

The Savanna Tree AI baseline dataset is published at Zenodo. The mosaics are also are published at Geonadir.

Setting up your desktop

Follow the steps below to install this repository on your dekstop.

• Install Azure Storage Explorer
• Install Anaconda
  • Choose All Users
• Install Git Desktop
• Open Git Desktop and clone https://github.com/ajansenn/Kakadutree.git
• Open Anaconda Prompt, cd to folder with cloned repository and [create conda environment] using the code below(https://docs.microsoft.com/en-us/azure/machine-learning/service/how-to-configure-environment#local):

conda env create -f environment.yml
conda env list
conda activate savannatreeai
conda install notebook ipykernel
ipython kernel install --user

Image Classification with Azure Custom Vision

Download the mosaic and shape files from Zenodo. Use the Raster to Custom Vision Jupyter notebook to clip images at the dimension of the bounding box and upload images (with labels) to Azure Custom Vision service.

CustomVision.ai in Microsoft Azure is used for image labeling and model training.

• Create a CustomVision.ai project

In Anaconda:

• Change directory to savannatreeai:

cd savannatreeai

• Start the Jupyter Notebook app:

jupyter notebook

Object detection and instance segmentation with Azure Machine Learning

Download the 1024x1024 training and validation datasets from Zenodo.

Deploy notebooks and dataset in Azure Machine Learning Studio.

Contact

For more information contact:

• Andrew Jansen, Supervising Scientist
• Steve van Bodegraven, Microsoft