Crop yield estimation is carried out in mango orchards
4–6 weeks prior to harvest to predict resource requirements for
the harvest and to arrange marketing. Estimation is a manual process, sometimes aided by the use of hand counters. Typically a few
‘sentinel’ trees are assessed per block, but nonetheless many man
hours are required in field conditions of heat and humidity. Ideally,
load would be assessed at several times during crop growth but labour requirements prevent this. Further, the sampling procedure
limits the accuracy of the process, in terms of estimating the yield
of the whole orchard.
A machine vision based system could replace the manual system. Several sensor technologies hold promise for this application,
including LIDAR, thermal imaging and stereovision. However, the
simplest and lowest cost solution would involve 2D machine vision. A camera mounted to a farm vehicle that drives between tree
rows could be used to acquire images of all trees on the farm relatively quickly. An appropriate image processing system would
need to be robust in its identification of mango fruit (i.e. require
minimal input from farm staff). Further work is required to relate
the fruit count from two images of a tree (as seen from the interrow on each side) to the total fruit load of the tree, a relationship