High-Throughput System for the Early Quantification of Major Architectural Traits in Olive Breeding Trials Using UAV Images and OBIA Techniques

• Main Authors: Ana I. de Castro, Pilar Rallo, María Paz Suárez, Jorge Torres-Sánchez, Laura Casanova, Francisco M. Jiménez-Brenes, Ana Morales-Sillero, María Rocío Jiménez, Francisca López-Granados
• Format: Article
• Language: English
• Published: Frontiers Media S.A. 2019-11-01
• Series: Frontiers in Plant Science
• Subjects: remote sensing; unmanned aerial vehicle; table olive; breeding program; training system; tree crown area and volume
• Online Access: https://www.frontiersin.org/article/10.3389/fpls.2019.01472/full

The need for the olive farm modernization have encouraged the research of more efficient crop management strategies through cross-breeding programs to release new olive cultivars more suitable for mechanization and use in intensive orchards, with high quality production and resistance to biotic and abiotic stresses. The advancement of breeding programs are hampered by the lack of efficient phenotyping methods to quickly and accurately acquire crop traits such as morphological attributes (tree vigor and vegetative growth habits), which are key to identify desirable genotypes as early as possible. In this context, an UAV-based high-throughput system for olive breeding program applications was developed to extract tree traits in large-scale phenotyping studies under field conditions. The system consisted of UAV-flight configurations, in terms of flight altitude and image overlaps, and a novel, automatic, and accurate object-based image analysis (OBIA) algorithm based on point clouds, which was evaluated in two experimental trials in the framework of a table olive breeding program, with the aim to determine the earliest date for suitable quantifying of tree architectural traits. Two training systems (intensive and hedgerow) were evaluated at two very early stages of tree growth: 15 and 27 months after planting. Digital Terrain Models (DTMs) were automatically and accurately generated by the algorithm as well as every olive tree identified, independently of the training system and tree age. The architectural traits, specially tree height and crown area, were estimated with high accuracy in the second flight campaign, i.e. 27 months after planting. Differences in the quality of 3D crown reconstruction were found for the growth patterns derived from each training system. These key phenotyping traits could be used in several olive breeding programs, as well as to address some agronomical goals. In addition, this system is cost and time optimized, so that requested architectural traits could be provided in the same day as UAV flights. This high-throughput system may solve the actual bottleneck of plant phenotyping of “linking genotype and phenotype,” considered a major challenge for crop research in the 21st century, and bring forward the crucial time of decision making for breeders.