Abstract

The availability of georeferenced maps of individual trees and their characteristics such as position in the orchard, canopy area and vegetation indices provide valuable information about spatial variability and crop status for plant-scale precision fruticulture and viniculture. The OrchardMapper desktop application was developed to detect trees and derive individual tree variables from 2D orthomosaics created from images obtained using unmanned aerial vehicle (UAV) platforms. The application is currently being used as part of a commercial precision agriculture and fruit yield estimation service. Applications include mapping of spatial variability of tree vigour for precision canopy management, derivation of tree sampling protocols, tree inventories and detection of missing plants. The algorithm is general enough to permit detection of trees under a wide range of crown coverage structures, ranging from closed carpet-like canopies of overhead trellis systems, to almost- or fully- discrete canopies characteristic of fruit and nut tree orchards. It also estimates portions of the canopy under shading or overexposure (situations resulting in loss of information pixels). We present a general overview of the software and algorithm and evaluated detection performance for four orchard structures. The cases are for multispectral othomosaics of a table grape vineyard, a wine grape vineyard, an apple orchard and an almond orchard with different training systems producing different conditions of canopy closure. Tree numbers ranged from 1500 to 3900 trees in the orchard blocks. Detection of rows and their direction was achieved at close to 100% detection rate. Tree detection rates (counts and positions) were 98% or better.