About

The laboratory was organized in Skolkovo institue of Science and Technology (Skoltech) to develop and apply Machine learning algorithms for intelligent analysis of Earth observation data.
It was supported by the National Technological Initiative within the project led by the University of Innopolis "The Digital Model of Tatarstan".
We cooperate with several research groups within Skoltech and outside and are open for collaboration in the area of applied Machine learning, remote sensning data processing, designing of EO downstream applications etc.
Our main work in the project is to create algorithms for monitoring of protected areas to support decision makers in the operating of spatially distributed assets (see Industries).
With the power of Skoltech High Perfomance Computing facilities we are able to perform fast experiments on the large datasets of the remote sensing data and produce the implementation of the algorithms as a data processing pipelines.

Industries

Building & Construction

Fast dynamics of urban areas is enforced by investment decisions. Development markets can be surveyed using instance segmentation and Change Detection methods - to estimate and classify buildings and assess population. We apply photogrametry and indirect methods to reconstruct buildings heights and distinguish one type from another.

Forest monitoring

Deforestation and tree cover are the common targets for remote sensing. Our applied works include early detection and monitoring of wildfires, classification of tree losses. We cooperate with IoT Lab at Skoltech to propose new combined measurments from the set of various sensors. Our partnership with Planet Inc. allows to create and implement the unique data models for the monitoring of the forestry.

Agriculture

We use algorithms based on deep learning models for classification of crops and detection of anomalies in vegetation growth. Collaborating with industrial partners and data providers we implement ML into real business processes of the agriculture workflows.

Monitoring of pipeline infrastructure

We are involved in actively research of change detection and object recognition in satellite and aerial imagery according to risk mitigation and filed work optimization for pipeline operator companies.

Monitoring of powerlines

The use of various sources of commercial satellite data allows us to fuse data to achieve the maximum coverage at optimal rate. In the current research we are focusing on the overgrowing of the powerline glades for clearing tasks and control.

Research

Open spatial dataset

“Open Spatial Dataset” is a joint project of universities and data operators to provide RnD community with open data and benchmarks for ML analysis of Remote Sensing data.
The labeling data is specified according to wide classification range of the natural and man-made objects that have a clear interpretation either in satellite or in aerial imagery.

Geoalert Platform

Geoalert company is creating the geoinformation platform that powers automatic analysis of Earth Observation data by pretrained Machine learning algorithms. Using Geoalert platform customers will be able to setup tasks for object recognition and define a territory of interest to get analytic reports and notifications of changes based on the recent Earth Observation data.

See how it works in the applications:

We use data provided by the leading global satellite operators as well as data acquiered by local aerial sensning companies.

News

Earth Observation and Machine learning - the hype is over now it's gonna to be real value proposition
We led two seminars within the Space center introductory course "Fundamentals of Remote sensning".
The next serie is going to be organized in the beginnig of summer, if you are interested to take part - please contact us.
4.0
New model for Forest detection and monitoring
Within the project of "Digital Model of Tatarstan" the new model for forest segmentation was created. The model will be used to detect such phenomena as overgrown of glades for protected areas of powerlines. It's adjustable for different resolution and was tested particular on SPOT imagery (1.5m) with NIR - to be applied to the SPOT mosaic of the whole territory of Tatarstan.
3.0
Open dataset service / May 19, 2018
We started publishing first datasets of the "Open spatial dataset" - the catalog of labeled satellite/aerial imagery for machine learning applications.
The one based on the Digitalglobe satellite imagery is aimed for Emergency mapping applications to detect changes in urban infrastructure in the aftermath of disasters. It contains the baseline Model for instance segmentation of the damaged houses in the residential areas of California region.
For dataset please visit our github page.
1.0