Deep learning for crop load management

Abstract

Effective crop load management in orchards is a requirement for accurate crop yield estimation. Traditionally this involves various methods of obtaining a measure of the fruit tree features that predominantly determine crop yield (wood, buds, flowers, fruitlets, and fruit). Manual counting of fruit, flowers, or fruitlets during various stages of growth is a laborious and expensive process and often suffers from significant inaccuracies. While optical approaches to automated fruit counting have been proposed they are usually not robust to changing light conditions or colour and shape changes of the fruit during the growing season.