sen2classification 0.0.1

Python library for automatic land cover classification of Sentinel 2 satellite images

sen2classification

sen2classification is a Python library for automatic land cover classification of Sentinel 2 satellite images using machine learning.

Features

• support for L1C and L2A Sentinel products
• uses machine learning algorithms in combination with tresholding for classifying land covers
• calculates spectral indices used for tresholding
• Two classification methods are available:
  • ALCC (Automatic Land Cover Classification)
  • Gradient boosting
• the end result is a classification raster which pixel values range from 0 to 5:
  • 0 = no data
  • 1 = water
  • 2 = low vegetation
  • 3 = high vegetation
  • 4 = soil
  • 5 = built up

Clouds are not classified, for optimal results use images with none or minimal cloud cover

ALCC

Automatic Land Cover Classification algorithm uses k-means clustering and tresholding for classifying land cover. The concept of ALCC was first introduced by Gašparović et al. (2019). They used it to classify Landsat-8 images at 30 m spatial resolution. This version of ALCC algorithm is adapted for use on Sentinel 2 images with 10 m spatial resolution.

Gradient boosting

Gradient boosting classification is implemented using xboost library. Collection of training data is automated by tresholding spectral indices.

Installation

Important!

Library dependency that must be installed before installing sen2classification is rasterio (and GDAL which is rasterio dependency). Installing sen2classification won't install rasterio.

Simplest way to install GDAL and rasterio is by using conda with conda-forge channel

conda install -c conda-forge gdal rasterio

After that use pip to install sen2classification.

pip install sen2classification

Usage

from sen2classification import ALCC
from sen2classification import GBClassification

# input is Sentinel 2 L2A product folder
alcc = ALCC(product='path to unzipped Sentinel 2 L2A product (.SAFE)', product_type='L2A')
# input is Sentinel 2 L1C product folder
xgb = GBClassification(product='path to unzipped Sentinel 2 L1C product (.SAFE)', product_type='L1C')


# sets up band file paths
alcc.setup_bands()

# swir bands must be resampled to 10m for calculating spectral indices
alcc.resample_swir_to_10m()

# calculate AWEI, NDVI, NDTI and BAEI spectra indices
alcc.calculate_spectral_indices()

# run ALCC algorithm
# final classification raster is saved in current working directory
alcc.run()

# setup and execution is identical for xgb

License

MIT