Smart India Hackathon 2020

Automated classification of land use using AI/ML

Develop a deep-learning based software for automatically classifying land-use from multi-temporal multi-spectral high-resolution satellite imagery. The developed model should be scalable/efficient to allow rapid mapping of incoming datasets and must incorporate a web-based viewer for visualizing input as well as classified output. The viewer interface must also allow the user to visualize changes that have occurred within a given timeframe.

NM386

Automating Land-use analysis

Using state of the art AI/ML techniques for analyzing land-use data.

Gathering Satellite Image

For showcasing our problem statement we decide to use the LISS-IV immages of the Republic of Slovenia. This image data was freely available on Sentinel Hub provide under a Creative Commons License.

The images covered an area of 250km X 170km and was split into approximately 300 boxes each of approximately 1000px X 1000px.

Each created box(an EOPatch) was then filled with image data collected from Sentinel Hub S3 buckets.

Building Reference Maps

The Machine Learing approach used in this case uses a Reference vector data to classify the land use in a specified Area of interest

The required refernce data was available in the form of shapefiles in the official sites of Republic of Slovenia

This reference shapefile with its organizational data is used to create a multi-dimensional volume of training data for the LBGM classifier.