BenthicClassifier

Classification of benthic point clouds

Step 1: Generating labels

We need to generate Hackel features for point cloud data using, and clean lines with NaN:

cat xyz_text_file | ./soundings_generate_features radius >> outputfile.Hackel

Example radius values:

• For MBES: 10
• For LiDAR: 100

or process a MBES batch

bash script/hackel_batch.bash directory outputDirectory radius

Step 2: Matching labels with ground truthing

Use the generate-training-data.py script to generate labeled training data:

python3 src/generate-training-data.py radius path-to-dfo-labels.csv directory-containing-dot-Hackel-files > training-data.txt

Example radius:

• For MBES: 3
• For LiDAR: 10

Step 3: Training the model

Using the traning data text file, train the model:

python3 src/train-model.py training-data.txt

This will generate a trained.model binary file that can be loaded with pickle in Python scripts or used in the following section

Step 4: Using the model

1. Generate features for the new files with soundings_generate_features:

2. Use the apply.py script to apply the model to a new text file:

python3 src/apply-model.py trained.model new_xyz_file.Hackel

BGX vs GMM

Step 1: Train both model

python3 src/apply-model.py trained.model new_xyz_file.Hackel
python3 src/gmm-best-fit.py new_xyz_file.Hackel max_nb_class

Step 2: Apply model

python3 src/apply-both-model.py model-file gmm-model-file hackelDataFile > outputClassification.csv

or in batch

bash script/classify-mbes-data.bash hackelDirectory

Step 3: Generate a comparative table and 3D bar plot

python3 src/compare-model.py classifiedData

or in batch

bash script/gen_table_figure.bash classificationsDirectory

Step 4: Generate raster

bash script/genraster.bash classificationsDirectory resolutionX resolutionY epsgCode