/tropical_forest_gunshot_classifier

Primary LanguageJupyter Notebook

Tropical forest gunshot classifier

Reproducability code for Katsis, L. K. D., et al. (2022). "Automated detection of gunshots in tropical forests using convolutional neural networks." Ecological Indicators 141 doi: https://doi.org/10.1016/j.ecolind.2022.109128

This associated data used for model training and validation (749 gunshots, and 35239 background sounds collected in tropical forest sites in Belize) is archived on Mendeley Data doi: https://doi.org/10.17632/x48cwz364j.3

For detailed information and instructions on how to run predictions using this model, see the Opensoundscape docs.

Requirements:

  • Python 3.7
  • Opensoundscape

Quick start

Classify files using the Google Colab script here

  • Opensoundscape 0.7.1.

Gunshot classification GUI

  • User-friendly scripts to run the classification algorithm on your data are found in the subfolder Gunshot classification GUI
  • This page contains instructions on running the scrips, and example data.
  • Opensoundscape 0.7.1

Model training and classification

  • Scripts used to reproduce the manuscript are found in the subfolder Model and code
  • This folder contains the training script, the final model, and the prediction script
  • IMPORTANT - this model currently must be opened within Opensoundscape 0.5.0 or 0.6.0 to replicate results in paper. Results differ if opened in later versions of opensoundscape