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

Purpose and goal

Deep Forestry has developed an autonomous drone that will produce vast quantities of digitized forest point clouds months before any harvests occur. This project will further develop an existing 3D AI algorithm that can semantically classify this abundant new drone-based data source in order to replace previously used guesswork practices. The project will optimize the Deep Forestry AI in an effort to help automatize downstream biomass transport, sawmill operations, and value chain logistics for commercial forestry.

Expected results and effects

The project aims to improve the forestry-specific classification capacity of the existing Deep Forestry AI using novel new deep learning tools applied in collaboration with Linköping Universities computer vision department. The optimized AI will be tested by real world users in an effort to directly demonstrate the added value of AI segmented 3D maps throughout various levels of the commercial forestry value chain. The success of the project will produce an AI tool that can be scaled up to optimize biomass production and removal methods throughout the commercial forestry value chain.

Planned approach and implementation

The project is split into four work packages: (1) 3D AI optimization and improvement (LiU). Various state-of-the-art methods will be applied to improve the existing Deep Forestry 3D AI algorithm. (2) LiDAR data acquisition(SCA). SCA´s fieldworkers will gather the data that is required to test and improve the AI algorithm. (3) Value chain improvement (SCA). SCA will use the segmented AI output to implement and asses the usefulness of the data for value chain improvements in commercial forestry. (4) Project administration (DF). Deep Forestry will administer the project.

External links

This webpage is currently an "under construction" placeholder. We expect to update the page sometime in 2021.