Desktop Site (Beta)

Introductory Video

Satellite data is for everyone: insights into modern remote sensing research with open data and Python

Loading

Follow to receive video recommendations   a   A
Speaker: Are you the speaker?

The largest earth observation programme Copernicus(http:--copernicus.eu) makes it possible to perform terrestrialobservations providing data for all kinds of purposes. One importantobjective is to monitor the land-use and land-cover changes with theSentinel-2 satellite mission. These satellites measure the sunreflectance on the earth surface with multispectral cameras (13 channelsbetween 440 nm to 2190 nm). Machine learning techniques likeconvolutional neural networks (CNN) are able to learn the link betweenthe satellite image (spectrum) and the ground truth (land use class). Inthis talk, we give an overview about the state-of-the-art land-useclassification with CNNs based on an open dataset.The EuroSAT benchmark dataset (http:--madm.dfki.de-downloads) is freelyprovided by German Research Center for Artificial Intelligence (DFKI).It consists of 27.000 image patches for ten different land use-coverclasses, e.g. industrial and residential areas, different crop andvegetation types and forests. All samples have 64 by 64 pixel dimensionand include either only the RGB images or all 13 bands.We will use different out-of-box CNNs for the Keras deep learninglibrary (https:--keras.io-). All networks are either included in Kerasitself or are available from Github repositories. We will show theprocess of transfer learning for the RGB datasets. Furthermore, theminimal changes required to apply commonly used CNNs to multispectraldata are demonstrated. Thus, the interested audience will be able toperform their own classification of remote sensing data within a veryshort time. Results of different network structures are visuallycompared. Especially the differences of transfer learning and learningfrom scratch are demonstrated. This also includes the amount ofnecessary training epochs, progress of training and validation error andvisual comparison of the results of the trained networks.Finally, we give a quick overview about the current research topicsincluding recurrent neural networks for spatio-temporal land-useclassification and further applications of multi- and hyperspectraldata, e.g. for the estimation of water parameters and soilcharacteristics. Additionally, we provide links to the code and datasetused in this talk.

Editors Note:

If you like this website, please upvote my Awesome Python pull request.

Comment On Twitter