sitsfeats.py

sitsfeats.py extracts features (metrics) from satellite image time series, with operations implemented in C++ for performance (Eigen) and a small, ergonomic Python API that integrates with the scientific Python stack.

This project is a modern re-implementation of the metrics from the stmetrics and sitsfeats (R) projects.

Installation

To install the sitsfeats.py package, please use the following command:

pip install sitsfeats

For more information on the options available for installation, please check the Installation page.

Quick example

Everything starts from a 2-D array where each row is one time series. The example below builds a thousand series of 36 observations each:

import numpy as np
from sitsfeats import feats

data = np.random.default_rng(0).normal(5000, 800, size=(1000, 36))

data.shape
# (1000, 36)

Now ask for the metrics you want, by name:

result = feats(data, ["mean", "skew", "amplitude"])

The result carries the metric names and a stacked matrix, one row per series and one column per metric:

result.names
# ['mean', 'skew', 'amplitude']

result.data.shape
# (1000, 3)

Column j always corresponds to result.names[j], so the three metrics of the first series read like this:

result.data[0]
# array([ 4.9131e+03, -3.4612e-01,  2.9532e+03])

See Installation to get started, Usage for the API and the data cube support, and the Metrics catalogue for every available metric and its definition.