3D Machine Vision for Improved Apple Crop Load Estimation

Karkee*, M.; Zhang, Q.; Lewis, K.M.

Accurate estimation of apple crop load is essential for efficient orchard management. In this work,
we designed an over the row platform to capture images from two side of apple canopies to
minimize the occlusions and 3D Machine Vision for Improved Apple Crop Load Estimationimprove the accuracy of crop load
estimation. A color camera, a 3D camera and a orientation sensor were
mounted in the sensor platform and moved down apple rows in three
different commercial orchards of Allan Bros. Inc., Prosser, WA.
Overall, the images of apples trees were successfully captured from
both sides of the row using this platform. Taking images from dual
sides showed to be fruitful as more apples were identified which were
occluded when viewed from a single side. A apple identification
algorithm and a 3D mapping algorithm was used to count apples while
avoiding duplicate counting of apples that were visible from both sides
of tree canopies. Crop load estimation improved by approximately
20% when imaged from two sides compared to that with single-side
imaging. Increased accuracy of crop load estimation on a block by block basis would lead to
increased efficiencies, a higher level of risk management and increased profitability.



If you would like more information about this topic or other developments in agriculture, contact Dr. Qin Zhang.