Sensing Technology for Automation in Forestry

Kengo Usui (Department of Forest Engineering)

Research Field: Automation in forestry

Main subject: Environmental sensing and recognition with images for autonomous forest machinery.

The Rreason that I Choose this Theme

By introducing new technologies, we can expand the operations of automated forestry. In the field of image sensing, there was a breakthrough in 2012: deep learning. This method enables end-to-end training, whereas the conventional method requires feature extraction to be carried out manually. This technology and its applications are now widely used in robotics, but their use in forestry is a work in progress. Previous research on the automation of forestry has been mainly focused on the driving of forest machinery; with deep learning, we can extend autonomous operations to include harvesting, loading, and cable yarding.

The Main Result

For the autonomous driving and working of forest machinery on a strip road in forests, localization and path planning are essential technologies. Adapting to dynamic environments in the forest, I have developed a road detection method to plan the appropriate path.
This system detects strip roads by semantic segmentation using deep learning. Deep learning usually requires a lot of training data.
Instead of preparing a huge training dataset, we generated images by the other type of deep learning. Figure 1 shows images of strip roads generated by deep learning. This image augmentation method enabled training of the deep learning model from only a small dataset.

Fig. Images of strip road generated by deep learning.

For more details, please see the paper below.

Usui K. 2021. Data augmentation using image-to-image translation for detecting forest strip roads based on deep learning. Int J For Eng. 32(1):57–66. https://doi.org/10.1080/14942119.2021.1831426

Final Aim of Research

Our aim is the development of completely autonomous machinery.

The first goal is commercialized autonomous driving for forestry machinery within 10 years. The important point is that automation should focus on human well-being, not be just a replacement of human labor. To define human well-being technology, we have to define the desirable forest. As the target forest changes, the operations to be automated will be also changing.

International Collaboration

There are not enough researchers for autonomous systems in the forestry sector because there are currently so many forest operations to automate. These systems to automate operations are versatile and could be used anywhere in the world, we still need to develop various technologies to overcome a few shortcomings. By developing technologies that can be shared, we can advance autonomous technologies on an international scale.