Model Agnostic Meta Learning (MAML) for 3D Forestry Artificial Intelligence

Purpose and goal

Deep Forestry, SCA, and LiU have together undertaken the work to produce and test a new class-incremental learning method in order to facilitate efficient expansion of the set of known classes to handle new tasks and environments. The method assisted in the roll-out of a commercially available artificial intelligence algorithm used for the classification of forest inventory, species, biodiversity, and ecosystems. Key forestry value chain improvements include databasing of all harvest age wood on an individual tree level and digitization of harvester decision making.

Expected results and effects

The project assisted in the roll-out of a commercially available AI algorithm that can be used for the classification of forest inventory, species, biodiversity, and ecosystems. This AI system was sold as part of multiple international pilot studies generating income for Deep Forestry. The project lays the foundation for an AI system that can segment and classify forest ecosystems globally. Recommendations about forestry value chain improvements were used as a guide to further develop, improve, and scale, the Deep Forestry product line, and to help close an investment round.

Planned approach and implementation

The implementation of the project included some adaptations. After the initial phases of the project it was eventually decided that Deep Forestry would take the lead role of labeling the training data. This occurred successfully and the commercial version of the AI was trained and validated with real world data by the end of the project. Following project completion Deep Forestry AB has begun implementation of the class agnostic AI algorithm into their globally available cloud platform. The Deep Forestry AI has also been successfully trained on new classes since the project completed.

External links

The website for the commercialization of the project results.