Lichen Image Classification

This project is a CNN pipeline for classifying lichens in images.

Project Structure

config - configuration file that contains all the hyperparameters and paths to the data directories

data_prep - main script for downloading and cleaning the iNaturalist observation data

split - main script for splitting the image into test, training and validation sets

train - main script for training the model using the training and validation sets

utils modules

scraping - fetching images from iNaturalist and creating training, validation and test sets

obs_data - data wrangling functions for observation data, loading and label functions for images

plotting - plotting functions for observation and image data

modeling modules

cnn_model - convolutional neural network model architecture

evaluate - evaluating the trained model on a test set

Datasets

The raw data can be found in the data folder. It was downloaded using iNaturalist's export tool in two batches (before 2017-01-01 and between 2017-01-01 and 2025-04-19.)

Filters:

• Research grade
• Most identifiers agreed
• Open geoprivacy (location not obscured)
• United States

Since lichen are not restricted to one taxonomic group, 'lichen' were defined as observations matching the 1,017 genera on the Lichen genera wikipedia category page. Taxon IDs were queried using iNaturalist's API and then passed into the export tool.

The exact query parameters are listed here.