ISPRS 2020 paper : Can SPOT-6/7 CNN semantic segmentation improve SENTINEL-2 basedland cover products? Sensor assessment & fusion

This page present inference results in new format for a better understanding of our statements and conclusions. Each of the 3 main comparisons have a dedicated tab containing :

Metrics bar charts of test dataset
Maps of train, validation and test datasets

Full camera-ready paper is available here.

Enhanced visualisation of inference results

Sensors assessment between SPOT-6/7 and SENTINEL-2 : 6 classes experiment

Metrics

Maps for fusion data is currently not available and will be online after quartine.

Sensors assessment between SPOT-6/7 and SENTINEL-2 : 18 classes experiment

Metrics

OSO map enhancing with CNN SPOT-6/7 fusion experiment

Metrics

Additional informations

Warnings

Please keep in mind that we used SPOT images with 1.5m and 6m pixel resolution respectively for panchromatique and RGB sensors. Google aerial images displayed here are 0.25m pixel resolution on RGB sensors. They where not involved into calculations and are only here to give a ground thruth to the reader and to improve results understanding.

The OSO map shown is a special inference processed only on the displayed SENTINEL-2 tile and thus may differ from the France-wise inference.

Usage

Use the border color of graph bar to get nomenclature color of the maps.

Graph : you can remove a set by clicking on its legend.
Maps : Opacity control on the bottom left impact all train/validation/test dataset of the given inference.

References

Inference maps where generated thanks to GDAL. Maps are displayed with Leaflet library and grouped layer control plugin (slightly modified by myself). Graph are genreted with Chartjs library.

OSO Map is available at OSO Distribution and is generated by CESBIO.

This work has been achieved by O.Stocker and A.Le Bris at LaSTIG of IGN