Customer Stories

Valued at more than $2 billion annually, the California citrus industry is a bedrock of the state's agriculture economy. Deeply rooted in that industry is the Lindcove Research and Extension Center (LREC), a part of the University of California Division of Agriculture and Natural Resources, which manages nearly 600 trees, mostly citrus varieties. To continue to support critical citrus research, LREC managers continually look for effective technological management tools. One researcher used that interest to test the feasibility of integrating new deep-learning algorithms into object-based image analysis (OBIA) software to automatically identify multi-age citrus trees from UAS imagery. The successful first-ever test is sowing the seeds for a viable new approach to citrus tree management. overview Location CALIFORNIA TRANSFORMING THE WAY THE WORLD WORKS TRANSFORMING THE WAY THE WORLD WORKS As part of its management improvement tools, the LREC created a reference tree database with the location and attribute data of 2,912 individual trees using 1-meter imagery from the US Department of Agriculture's National Aerial Imagery Program (NAIP). In 2017, they became interested in UAS imagery. At the same time, Ovidiu Csillik, a visiting research scholar at UC Berkeley's Department of Environmental Science, Policy, and Management, and his research supervisor Dr. Maggi Kelly, were interested in how UAS imagery, a convolutional neural network (CNN) deep- learning algorithm and Trimble's eCognition OBIA technology could be used to accurately map multi-age citrus trees. "This study would give us the chance to acquire UAS data as a proof-of-concept alternative to NAIP imagery, and to test the feasibility of using a CNN-OBIA method to automatically analyze the UAS imagery and accurately identify and map individual citrus trees," said Csillik. In January 2017, researchers acquired UAS imagery of the entire 175-acre LREC site. In two flights, the UAS captured 4,574 multispectral images, which were photogrammetrically processed to produce a 4-band orthomosaic with a ground sample distance of 12.8 cm. The red and near infrared bands were also used to create a Normalized Difference Vegetation Index (NDVI) image. Both datasets were used as source data for eCognition. The first step of the ruleset involved training the CNN model with three classes—trees, bare soil and weeds—with 4,000 training samples per class. To create the samples, eCognition used the orthomosaic and the LREC's tree-location database to break up the mosaic into 40x40 pixel samples around

• Cover