Abstract
To optimize fruit production, a portion of the flowers and fruitlets of appletrees must be removed early in the growing season. The proportion to be removedis determined by the bloom intensity, i.e., the number of flowers present inthe orchard. Several automated computer vision systems have been proposed toestimate bloom intensity, but their overall performance is still far fromsatisfactory even in relatively controlled environments. With the goal ofdevising a technique for flower identification which is robust to clutter andto changes in illumination, this paper presents a method in which a pre-trainedconvolutional neural network is fine-tuned to become specially sensitive toflowers. Experimental results on a challenging dataset demonstrate that ourmethod significantly outperforms three approaches that represent the state ofthe art in flower detection, with recall and precision rates higher than$90\%$. Moreover, a performance assessment on three additional datasetspreviously unseen by the network, which consist of different flower species andwere acquired under different conditions, reveals that the proposed methodhighly surpasses baseline approaches in terms of generalization capability.