The goal of this study was to create an automated workflow where researchers can input lichen micro-plots photos and a Lichen coverage % would be calculated. We worked together to design and train a U-net based Convolutional Neural Network in TensorFlow. Our neural network, LiCNN, has an accuracy of 91%, with a dice coefficient of .92 with our Labrador dataset.
Link to paper: https://www.tandfonline.com/doi/full/10.1080/07038992.2022.2144179
After training LiCNN, I worked on a script used to automate imagery processing. The output includes a .csv file with numerical data, along with output images such as the ones below.
Additional credit to the Canada Centre for Mapping and Earth Observation, Julie Lovitt, Wenjun Chen, and Krishan Rajaratnam. A github repo containing LiCNN can be found at https://github.com/CCRS-lichen/LiCNN_Models
