Seminar – Machine Learning with spatial data

Are you working with geospatial data (raster, vector, point cloud) and machine learning? Join our seminar to share and learn best practices, discuss and network. 


The aim of this seminar is to share best practices of Machine Learning with spatial data and to exchange experiences and ideas among spatial Machine Learning researchers and practicioners in Finland. This will also give us at CSC the opportunity to learn current and potential users.

The idea is to concentrate on the technical side of Machine Learning with spatial data:

– Data preparation,
– tools,
– solutions and
– challenges.
The only prerequisite for participation is to have an interest in Machine Learning with spatial data. We also encourage companies to participate to learn about the possibilities of LUMI supercomputer for spatial machine learning. Experience of using CSC services in not mandatory, but of course can be helpful. We want this to be an open and welcoming event; especially for asking questions (”there are no stupid questions”).

The seminar is open and free of charge to everyone interested. You will get the most out of this seminar by attending in person in Keilaniemi, but we will also provide the possibility to listen to the talks online.

The event will provide the possibility for researchers and machine learning practicioners to share and discuss their work. CSC specialists will host the meeting, briefly tell about CSC resources for Machine Learning with spatial data and be available for questions.

Approximate schedule:

12:00 – 12:25

  • Introduction, welcome, practicalities (Samantha Wittke, CSC)
  • Short introduction to machine learning at CSC (Mats Sjöberg, CSC)

12:25 – 13:05

  • Deep Learning based Semantic Segmentation Workflow for the QGIS Plugin EnMAP-Box (Leon-Friedrich Thomas, Uni Helsinki)
  • geo2ml: python package for converting geospatial data to ML formats (Janne Mäyrä, SYKE)
  • A New Approach for Feeding Multispectral Imagery into Convolutional Neural Networks Improved Classification of Seedlings (Mohammad Imangholiloo, University of Helsinki)

13:05 – 13:45

  • Hyperspectral superresolution and multi-image fusion for remote sensing forest studies (Jorma Laaksonen, Aalto University)
  • Detecting soil cover on arable land in winter from satellite remote sensing ( Maria Yli-Heikkilä, LUKE)
  • 3-D modeling of biodiversity and ecosystem services (Parvez Rana, LUKE)

13:45 – 14:20

  • Break

14:20 – 15:00

  • Vector ML – case Gaze-Aware Interactive Map System (Pyry Kettunen, FGI)
  • Monitoring Bark Beetle Damage: Technical Insights into UAS Image Classification with Deep Neural Networks (Emma Turkulainen, FGI)
  • Predicting tree species specific growing stock using ALS point clouds and a 3D-Convolutional Neural Network (Andras Balazs, LUKE)

15:00 -15:30

  • Mapping invasive plant species using airborne hyperspectral data and weakly supervised deep learning (Ilja Vuorinne, University of Helsinki)
  • Building detection from Lidar data using 3DUNet (Jere Raninen, NLS)
  • How the diversity of training data improves the 2D UNet building detection (Emilia Hattula, NLS)

~15:30 – 16:00

  • Discussion, Q&A