MISTRA DIGITAL FOREST ANNUAL REPORT 2020 1
FOREST REMOTE SENSING Our dream is to figure out
which scanning method should be used to obtain information about tree species, which method gives the best results and is most cost-effective JOHAN EKENSTEDT New project on AI forwarding – Tree species is something of a holy grail that we really want to discover, because if we don’t know what tree species we have, we don’t know what products we will get either. In addition to tree species, in order for it to be really great, what’s missing is knowing how many trees there are per hectare, he says. – If that could be solved, many time-consuming measurements on the ground could be avoided. With the work done in Mistra Digital Forest, Johan Ekenstedt hopes to produce data as good as – or better than – measurement in the field. – Our dream is to figure out which scanning method should be used to obtain information about tree species, which method gives the best results and is most cost-effective. It costs a good few millions to scan the forest, he says. Small trees – not less important It’s not just identifying the large trees that’s important, but also the lower layers of vegetation. Being able to see small trees is beneficial from a biodiversity perspective, small trees can act as cover for birds, for example. Also, from an economic perspective, the information in the lower vegetation is valuable for planning clearing and thinning with greater precision. Langning Huo, postdoctoral student at SLU, is conducting a study on lower vegetation layers and the analysis is almost complete. – The method Langning Huo uses clusters the points into groups that correspond to tree crowns or shrubs. Then, all the tree crowns in the top crown layer are identified and removed. In this way, you can obtain information about what is further down, Eva Lindberg concludes. Is it possible to teach a forwarder controlled by artificial intelligence (AI) to load logs in a digital environment and then transfer the ability to physical machines? Martin Servin and colleagues at Umeå University will investigate this question in a new project with funding from the programme's strategic reserve. The project is about autonomous forwarding, in particular the mechanical puzzle of gripping and loading logs in a safe, energy efficient way, says Martin Servin who leads the project. He continues; – We will investigate the possibility of making the AI "aware" of the physics of crane manipulation of logs by training it in a virtual environment with digital physics. Broad collaboration – from Umeå to California The project has multiple purposes and extensive collaboration. Participating actors include the Swedish University of Agricultural Sciences (SLU), Skogforsk and Holmen Skog – and more forest companies are welcome to participate. Collaboration is also planned with Luleå University, KTH and UC Berkeley in the USA. – Beside these central scientific questions, we are seeking greater understanding of the potential and the challenges of using artificial intelligence for automation of forest machines, and to stimulate the machine manufacturers to take new leaps in automation, says Martin Servin. Another aim is to establish collaboration with worldleading researchers in AI robotics (UC Berkeley). – They have taken a liking to our simulation technology, and we have a lot to learn from them. Through this project, we also want to get them interested in forestry robotisation, he concludes. 9