I am Miroslav. I work at Savantic AB and this is my PhD project website.

Why accelerating taxonomy?

If we strive to have a more complete inventory of life and preserve biodiversity for upcoming generations, then we need to accelerate taxonomic delivery. This involves making relevant information accessible to everyone.

How?

Images and machine learning

Machines powered by deep learning and CNNs are inspired by the way the human eye and brain work and they overperform the humans on many visual tasks.

Citizen Science

Research isn’t just for researchers. More than 80 % of the 430 million records from the Global Biodiversity Informatics Facility (GBIF) are from citizen scientists.

Automated taxon identification:

Imagine being able to identify insects by just photographing them with your smartphone!


additional info

It could link you to the sources with more information such as Naturalist or Wikipedia.


Collecting more biodiversity data:

When image is uploaded locality is stored and occurence is linked to GBIF.

Scientists with special interest in taxon could be instantly notified.

If you don't know who is on the image then right click and try searching Google for image.


Our goal

To apply cutting-edge image recognition technology to a widespread, diverse, but poorly-known insect groups.

Think "face recognition" but with insects instead of faces.

Missions

to be added as the project moves forward.

Beetle families

novel species

dorsal view

Workflow

Iteratively

each iteration consists of the following six steps

  • Step 1

    TTT - Target the taxa

    The primary source of specimens for the image acquisition are museums world wide (identified specimens) and SMTP material, which includes over 50 millions of specimens in alcohol, sorted into 50 taxonomic fractions by experts.

  • Step 2

    Collect the images

    High resolution images taken in lab settings (camera or microscope) but also smartphone images taken with or without attachable lenses.

  • Step 3

    Preprocess

    If inevitable, this step will be done using OpenCV, an open source library for Computer Vision

  • Step 4

    Train the machine

    To train the machine to recognise objects and details in the picture we need training data (images), powerful learning algorithms (DEEP LEARNING) and sufficient computational power (GPU).

  • Step 5

    Classify

    Classification involves dividing the data into various interrelated classes.

  • Step 6

    Verify

    To verify if the machine clasified images properly, identification is required for all the specimens used in the study.


  • repeat




    Step 0

    Find another tasks

    Test the model on new identification task.

>

Our Team

This project is part of Miroslav's PhD thesis.

Miroslav Valan

PhD student

Savantic AB
Stockholm University (SU)
Swedish Museum of Natural History (NRM)

Fredrik Ronquist

Professor

Swedish Museum of Natural History (NRM)

Atsuto Maki

Associate Professor

KTH Royal Institute of Technology

Karoly Makonyi

Researcher

Savantic AB
Stockholm University (SU)

This project is part of the BIG4 global consortium.

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No 642241.





Contact Us

If you find this interesting, have a question or would like to

contribute or colaborate

contact Miroslav directly via e-mail or via form below