Awesome

About

Code and data to implement Buscombe & Carini (2019) to classify wave breaking using deep neural networks

Buscombe and Carini (2019) A data-driven approach to classifying wave breaking in infrared imagery. Remote Sensing 11 (7), 859; https://doi.org/10.3390/rs11070859

https://www.mdpi.com/2072-4292/11/7/859

Software and data for training deep convolutional neural network models to classify wave breaker type from IR images of breaking waves in the surf zone

Folder structure

• \conda_env contains yml files for setting up conda environment
• \conf contains configuration files using for training models
• \train contains files using for training models
• \test contains files using for testing models
• \out contains output files from model training
• \keras_mods contains modified keras applications files

Setting up computing environments

These task requires modifications to keras libraries. The easiest way to deal with two different keras installs is to use conda environments

First, some conda housekeeping

conda clean --packages
conda update -n base conda

1. Create a new conda environment called classification

conda env create -f conda_env/classification.yml

C:\Users\ddb265\github_clones\IR_wavegauge\

2. Copy the contents of the keras_mods\classification\tf_python_keras_applications folder into the tensorflow\python\keras\applications site package in your new conda env path. For example:

C:\Users\user\AppData\Local\Continuum\anaconda3\envs\classification\Lib\site-packages\tensorflow\python\keras\applications

Be sure to keep a copy of the existing files there in case something goes wrong.

3. Copy the contents of the keras_mods\classification\tf_keras_applications folder into the tensorflow\keras\applications site package in your new conda env path. For example:

C:\Users\user\AppData\Local\Continuum\anaconda3\envs\classification\Lib\site-packages\tensorflow\keras\applications

4. Activate environment:

conda activate classification

Deactivate environment when finished:

conda deactivate

Training classification models

Extract image features

The following has been tested with the following models: MobileNetV1, MobileNetV2, Xception, InceptionV3, InceptionResnet2, and VGG19

1. Run the feature extractor using the MobileNetV2 model, with augmented images, running extract_features_imaug.py and the configuration file conf/conf_mobilenet.json:

python extract_CNN_features_imaug.py -c conf_mobilenet_imaug

To do the same thing using non-augmented imagery, use:

python extract_CNN_features.py -c conf_mobilenet

or augmented imagery with the xception model

python extract_CNN_features_imaug.py -c conf_xception_imaug

2. Run the feature extractor using the Xception model, without augmented images, running extract_features.py and the configuration file conf/conf_xception.json:

python extract_CNN_features.py -c conf_xception

Train and save model

Mobilenet, non-augmented:

python train_test_model.py -c conf_mobilenet

Mobilenet, augmented:

python train_test_model.py -c conf_mobilenet_imaug