Deep Learning Based Avocado Tree Counting from Drone Images on Chilean Orchards
Research paper on using deep learning and geometric modeling to automatically count avocado trees from drone imagery, helping Chilean orchards track tree loss from aging, soil compaction, and salinization.
Co-authored with Cameron Maloney, David Guerrero-Pantoja, Honeyman Andres, and Maria Pantoja. Avocado orchards lose trees gradually to aging, soil compaction, and salinization, and catching that loss early matters for yield planning. The approach combines Detectron2 for segmenting the orchard into regions, PCA and Projection Pursuit Optimization to figure out the dominant row orientation, and CountGD (a vision-language model) adapted to take advantage of how structured and linear orchard rows are, to detect and count trees from aerial imagery.