Computer vision is on the rise in forestry

Computer vision is a field of AI that enables the forestry machines of the future to understand, and to act on, visual information. Researchers have recently developed a fast and inexpensive data collection method, that lowers the threshold for the ability of the technology to really take off in the forestry sector.

AI is trained across many industries to analyse its surroundings using visual input. Computer vision, as it is known, is a technological development that could also prove extremely useful in forestry, as a means of identifying tree species and detecting tree damage, for example. What stands in the way of this development is the availability of annotated image data – that is, data that has been labelled so that a model can understand its content, and which AI can be trained on. There is an insufficient amount of this data, and collecting it manually is too time-consuming.

A project within Mistra Digital Forest is currently working on automating the collection of annotated image data. As a first step, the method involves collecting visual information using a camera mounted on the harvester, and then cross-referencing it with information from the harvester’s production file.

In this way, the image frame rates are matched with vast volumes of data on the characteristics and position of the tree, entirely without human intervention. In the next step, the researchers annotate the images manually. This semi-automated solution reduces processing time by 92%, compared with entirely manual methods.

– This is a quick and cost-effective way to collect, and to improve access to, training data, without disrupting the day-to-day forestry operations. The method is scalable and can be applied to different types of forests, machinery and situations. It paves the way for collecting data on a whole new scale, and it lays the foundation for models that improve safety and productivity in the forest in various ways, says Christian Höök, a PhD student at the Department of Forest Bioeconomy and Technology, SLU.

SCA: “Good both for the working environment and for efficiency” 

Based on the promising initial results, the researchers received further funding from Mistra Digital Forest’s strategic reserve. They will now automate the manual annotation process, and develop a hardware package consisting of a computer and a camera that will be mounted on the harvester. Preliminary tests are underway, and a hardware prototype is scheduled to be ready by the end of 2026.

For project partner SCA, forestry machines that see, observe and analyse their surroundings are of interest as a means of reducing the workload of the machine operator.

– Automating some of the machine operator's tasks would be beneficial both from a health and safety perspective, and in terms of efficiency. One potential first application is a computer-vision-equipped harvester that can automatically classify trees, with the aim of subsequently discovering various defects that affect timber value, such as warping. However, one prerequisite is that this type of solution is integrated into the machine. Once we see that this potential exists, we will most definitely be asking machine manufacturers for it, says Magnus Bergman, Head of Forestry Technology at SCA.