A methodology using AMP for estimating the yield of citrus
two weeks ahead of harvesting was introduced. The automatically
adjusted threshold method for image segmentation
suggested that it was able to obtain sufficient statistical
accuracy for citrus yield estimation. According to the characteristics
of the outline of the spherical fruit, a modified
8-connectedness chain code algorithm was used to determine
the number of fruits when they are clustered in the image. The
test result showed that the modified 8-connectedness chain
code algorithm contributed a satisfactory performance in
clustered fruit regions when the cluster contained less than
five fruit. Using the methodology for a single tree, a satisfactory
average estimation ratio (90%) between the actual and
predicted fruit yield was achieved. Thus, this study presents a
new method based on portable device for future citrus yield
prediction. Yield estimates of individual citrus trees obtained
by the developed methodology would provide valuable information
for planning harvest schedules and generating prescription
maps for site- and tree-specific application of
precision management. This method could perhaps be
applied for yield estimation of other crops whose fruit colour
shows obvious differences from their leaves.