lennijusten / tickidnet

A CNN-based approach for tick species identification.

TickIDNet

Justen, Lennart, et al. “Identification of Public Submitted Tick Images: A Neural Network Approach.” PLOS ONE, vol. 16, no. 12, Dec. 2021, p. e0260622. PLoS Journals, https://doi.org/10.1371/journal.pone.0260622.

1) Installation

You can install the required packages with conda and pip

Anaconda (recommended)

conda create --name venv python=3.6
conda activate venv
conda install tensorflow pillow numpy pandas

Virtualenv

pip install virtualenv
virtualenv .venv
source .venv/bin/activate
pip install -r requirements.txt

2) Making predictions

You can run TickIDNet on a batch of images with predict.py and the appropriate arguments in your terminal/console

python predict.py source dest model

where source is a path to a directory of images, dest is the path where you want to save the output CSV file, and model is the path to the saved model.

The script will internally crop any images in the directory (your files will remain unchanged) into a square along its shortest side and then resize it to the standard 224x224 input size. The tick will need to be near the center of the image and not, for example, on the upper edge of a portrait-style picture. See \Sample Image Data for examples.

There is also a strong correlation between the relative size of the tick in the image and the networks accuracy. For better results try cropping the pictures closely around the tick.

3) Model

The final model can be downloaded from Google Drive. See the docs for more info.

4) Data

Sample images available in this repository were used in the evaluation of TickIDNet. The majority of images used in the development of TickIDNet, including those in the \Sample Image Data folder, were initially sourced from iNaturalist. The citations below reference the publically available image data from iNaturalist:

Amblyomma americanum:
GBIF.org (16 July 2020) GBIF Occurrence Download https://doi.org/10.15468/dl.4gbcs6.

Dermacentor variabilis:
GBIF.org (16 July 2020) GBIF Occurrence Download https://doi.org/10.15468/dl.tyybke.

Ixodes scapularis:
GBIF.org (16 July 2020) GBIF Occurrence Download https://doi.org/10.15468/dl.sq29u5.

File naming conventions

The files in \Sample Image Data are all named in the following way:

Genus_species_sex_lifestage_source_alive_feedstage_#.jpg

For example: Dermacentor_variabilis_m_a_ta_unk_unfed_1.jpg

The labels have the following categories:
Genus: any
species: any
sex: male (m), female (f), unkown (unk)
lifestage: adult (a), nymph (n), larvae (l), unkown (unk)
source: Tick App (ta), WMEL Lab (MCEVBD), TickReport (tr), iNaturalist (iNat)
alive: dead (dead), alive (live), unkown (unk)
fed: fed (fed), unfed (unfed), unkown (unk)

tickidnet's Issues

Train data source

I can't find where is the data used for training, please can you provide it?