cnn_sea_turtle_detection

Code for Methods in Ecology and Evolution paper: "A Convolutional Neural Network for Detecting Sea Turtles in Drone Imagery"

Paper can be accessed at: https://doi.org/10.1111/2041-210X.13132

Using this code:

Running run.sh in bash will run the full turtle detection workflow.

Notes:

• Python 2.7 is required and nonstandard python packages necessary are: numpy, scipy, keras, tables, and hdf5storage
• This setup runs on preprocessed imagery contained in the .mat file. Full turtle image data along with labels for independent machine learning development can be found at doi:10.5061/dryad.5h06vv2

File Details:

• data.py
  • defines utility functions for model creation and matlab ingestion functions
• cnn_predict_stack.py
  • run prediction on processed images
• DukeTurtle_info.h5
  • Trained model weights file
• DukeTurtle_info.json
  • Model definition file
• DukeTurtle_test.mat
  • processed and tiled RGB image data that is fed into the model. Training / validation split is 85% train / 15% validation