---

NAME

m.neural_network - GRASS GIS addons to train and apply a neural network.

KEYWORDS

raster, vector

DESCRIPTION

The m.neural_network toolset consists of several modules.

• m.neural_network.preparedata_part1: Prepares and exports tiles for the label process
• m.neural_network.preparedata_part1.worker_export: Worker for parallel processing for exporting for m.neural_network.preparedata_part1
• m.neural_network.preparedata_part1.worker_nullcells: Worker to analyse the number of null cells in parallel for m.neural_network.preparedata_part1
• m.neural_network.preparedata_part2: Prepares imagery and labelled data for training and application of a neural network.
• m.neural_network.preparedata_part2: Worker to rasterize labelled data in parallel for m.neural_network.preparedata_part2
• m.neural_network.train: training of a semantic segmentation model with smp libraries
• m.neural_nework.test: calculation of statistics for quality assessment
• m.neural_network.apply: application of a trained model to new data

REQUIREMENTS

The following Python libraries are needed.

• grass-gis-helpers>=2.2.0
• GDAL/OGR and Python bindings
• pytorch
• segmentation_models.pytorch
• pytorch-lightning
• albumentations

AUTHORS

Anika Weinmann, mundialis GmbH & Co. KG, weinmann at mundialis.de

Guido Riembauer, mundialis GmbH & Co. KG, riembauer at mundialis.de

Markus Metz, mundialis GmbH & Co. KG, metz at mundialis.de

Victoria-Leandra Brunn, mundialis GmbH & Co. KG, brunn at mundialis.de