Currently, the European spruce bark beetle is one of the biggest risks to spruce forests in Europe. The human eye cannot detect the spread of the harmful beetles so easily, but AI is being trained at the University of Jyväskylä to spot the damages from images produced by a spectral imager. 

An impressive three-dimensional model rotates on a computer screen at the University of Jyväskylä’s Spectral Imaging Laboratory. The virtual forest looks like a normal forest with its various tree species, undergrowth, and undulating ground.

The virtual model of a forest was created by the Faculty of Information Technology. It must be as compatible as possible with a real forest because it is used to prevent the damages caused by the European spruce bark beetle.

Currently, the European spruce bark beetle is one of the biggest risks to spruce forests in Europe. The spruce bark beetles create corridors and lay their eggs under the spruce bark, provoking the death of the trees.

There has been a growing realisation of these risks also in Finland: The Finnish Forest Centre recently announced that they will contact the forest owners in risk areas and advised everyone to keep an eye out for signs of insect damage.

“For example, in some places in Central Europe and Czech Republic thousands of hectares have been cut down to stop the spread of beetles,” says Associate Professor Ilkka Pölönen, who is the head of the Spectral Imagining Laboratory at the University of Jyväskylä. “Climate change has caused for the beetles to spread hundreds of kilometres from southern Finland to the nothern parts in ten years.”

“The human eye cannot easily detect the proliferation of beetles,” says Pölönen.

“Or at least, not until it’s too late. It might be possible, however, to detect them automatically with the help of AI.”

Images of the forest are analysed using artificial intelligence.

Images of the forest are analysed using artificial intelligence.

AI assists in interpreting data captured by drones

However, in order to function, AI requires a lot of training material. The aim is to utilise the virtual forest particularly for the production of the training material for AI.

In the study, the detection of beetles is done with a spectral imager attached to a drone. The spectral imager takes images of the forest which are then analysed with the help of AI.  The objective is to find signs of damage caused by the European spruce bark beetle.

“It would take an awful lot of time to go through spectral images and detect the spread of the beetles, so it therefore makes sense for AI to perform the analysis,” explains Pölönen.

AI processes many things a lot more efficiently and quicker than humans do, but to function properly it requires enormous volumes of data.

“Collecting such a large amount of data in the real world would take an awful lot of time and resources,” says Pölönen.

“That would mean hundreds and hundreds of extra hours spent on fieldwork. However, the same amount of data can be created with the help of a virtual forest.”

It makes no difference to AI if we use a spectral imager attached to a real drone to take pictures of a real forest, or if we use a virtual drone, a spectral imager, and a forest. The main thing is that there is data available. Conducting AI research without a significant amount of data is vastly more difficult.

Picture of virtual forest.

Data obtained from the real world has been used for the modelling of the virtual forest to make the forest as realistic as possible. A small amount of data from the real world can be used to create a considerable amount of virtual data. Credit: Kimmo Riihiaho

The spectral imager is more efficient than a traditional camera or the human eye

The spectral imager used for capturing the damage caused by beetles is significantly more accurate than a regular camera, and it’s able to distinguish a much wider spectrum of colours. This makes it possible to spot things in the images that otherwise would go undetected.

“If you think about the image pixel on the computer screen, for example, one image pixel is made up of three different layers: red, green, and blue,” says researcher Kimmo Riihiaho from the University of Jyväskylä, who utilises spectral imaging. “Whereas a traditional image has three colour channels, a spectral image can have up to hundreds of channels depending on the requirements of the application.”

The huge number of narrow channels makes even the assessment of the molecular composition of the target of the imaging possible with the help of a spectral image.

A picture of spectral data.

This spectral data has been obtained from birch leaves measured in the laboratory. 

Kimmo Riihiaho is the man behind the three-dimensional virtual forest mentioned previously.  Next, Riihiaho opens a graph on the screen, which has been created from one single pixel of a spectral image.

A graph doesn’t mean anything to a human, but for AI it can be very useful. Spectral images contain such a vast amount of data that there is little likelihood for humans to start analysing them without machines.

“Once we obtain enough material on both healthy spruce trees and trees infected by the European spruce bark beetle, AI is able to automatically detect from the images which forest plots are potentially infected by the bark beetles as well as how widely the invasion of beetles has spread,” says Riihiaho.

Data similar to that of a real forest can be obtained from a virtual forest for the use of imagers. The model is able, for example, to adjust the amount of light reflected from leaves or pine needles and change countless amounts of parameters related to the leaves in such a way that the virtual trees are able to imitate the infected trees.

The contours of the terrain can be created according to the terrain map to match the existing forest if so desired.

“Our aim is to create virtually the kind of training material that can be used in the future to automate, for example, the monitoring of the spread of the European spruce bark beetle and to identify potential populations at an early stage,” says Ilkka Pölönen about the research objectives.

Researcher Kimmo Riihiaho in the forest.

Images taken from a real forest form the basis of a virtual forest that researcher Kimmo Riihiaho has built to train artificial intelligence.

Imaging to assist in disaster prevention

If detected early on, infected trees can be felled and thus prevent the spread of beetles more widely.

Ilkka says that with global warming, Finland’s beetle damage will increase and if the forests are not cut down then the beetles will eventually destroy them completely. It is not only an economic but also an ecological disaster if a 100-year-old spruce forest dies suddenly. Currently, the closest beetle infestations in Jyväskylä can be found in Laajavuori.

With the help of the virtual forest, AI and the spectral imager attached to the drone, the research laboratorium’s staff, however, hopes that that disasters can be prevented in advance.

One of the easiest ways to prevent damage to forests is to grow more heterogenious trees. The European spruce bark beetle only eats spruce trees.

“In the future, we must focus on combating disasters and discovering beetle populations as well as on how and what type of forests are grown in Finland,” explains Ilkka Pölönen.

Associate Professor Ilkka Pölönen leads the Spectral Imaging Laboratory at the University of Jyväskylä. Hyperspectral imaging can be used in forest management, remote sensing, agriculture and marine biology applications.

The Spectral Imaging Laboratory is part of the Faculty of Information Technology

Ilkka Pölönen and Kimmo Riihiaho are researchers at the Spectral Imaging Laboratory of the University of Jyväskylä at the Faculty of Information Technology. The laboratory uses spectral imaging with image data analysis.

The aforementioned study was a part of the ASPECT project financed by the Academy of Finland, which extensively examined the use of drones and AI in the assessment of the condition of forests. Research continues in the Academy of Finland’s ML4DRONE project, which starts in autumn 2023.

The studies were carried out by Professor Eija Honkavaara and the research group from the Finnish Geospatial Research Institute of the National Land Survey of Finland.

In addition to preventing beetle infections, the virtual forest can be used to train AI to detect, for example, the amount of nitrogen in the forest floor or the amount of water in the leaves of the trees in the spectral images. This way, the forest can be helped, for example, to grow faster.

The website for the Spectral Imaging Laboratory





Get latest articles from The University of Jyväskylä’s stakeholder magazine into your email. You can cancel your subscription at any time.