farmB developed a methodology able to identify tree canopies from aerial images. Based on AI algorithms, the system is capable of separating the trees’ canopies from the ground, providing clear canopy outlines for all trees located within an orchard.
Deep learning algorithm U-net, is utilized for providing semantic segmentation in images, taken from unmanned aerial vehicles. Tree center localization is performed based on each canopy’s weighted average, offering an accurate prediction on the location of each tree trunk.
A large number of orchards was used for development and testing, covering all seasons, tree development phase and weed presence. Aim was to create a robust system that is able to perform the identification task under variable open-air orchard environment conditions.
The insightful outcomes demonstrated the soundness of the approach under different conditions. The method offers real-time predictions and utilizes inexpensive equipment, signifying its ease-of-use for deployment in operational environments. The results provide canopy sizes for age estimation, pesticide application and yield prediction, as well as accurate mapping of orchards for path planning tasks.