Case Study: Carnegie Institution for Science achieves 96%‑accurate automated citrus tree mapping with Trimble eCognition

Combining UAS data and deep-learning OBIA technology offers new approach to citrus tree management

Lindcove Research and Extension Center (LREC), part of the University of California Division of Agriculture and Natural Resources, sought a faster, more precise way to map and manage multi‑age citrus trees across California’s $2+ billion citrus industry. The challenge was to test whether high‑resolution UAS imagery combined with deep‑learning and object‑based image analysis (OBIA) could automatically and reliably identify individual citrus trees for long‑term crop management.

Using two UAS flights that produced 4,574 multispectral images (12.8 cm GSD) and an NDVI layer, researchers trained a convolutional neural network inside Trimble’s eCognition OBIA with three classes (trees, bare soil, weeds). A sliding‑window CNN produced a probability heat map, and superpixel segmentation refined tree delineation. The workflow identified and delineated 3,105 trees in 30 minutes with 96.2% accuracy; LREC plans to test transferability and update maps over time.

Carnegie Institution for Science

Ovidiu Csillik

Postdoctoral Research Associate