Apple Growth Analysis Using Deep Learning Approach in Orchards

Abstract

Detection of apples in orchards during its growth can help in estimating the productivity, but detecting all the apples will be a challenging part as some apples might be very small and occluded by leaves and branches. Although deep learning-based image segmentation algorithms have shown successful outcomes in crop area delineation, this method is still unable to precisely segment the regions of every target apple with significant overlap. Region Proposal Networks like Faster R-CNN can be used for detection, but they are not efficient in producing better results when the apples are very small. Furthermore, these systems can only detect apples at a specific stage of development, but they can’t predict yield without first learning about the growth features of apples as they mature. In order to solve the above mentioned problems that are involved during apple detection in orchards, an enhanced version of the You Only Look Once(YOLO)-V3 model is proposed for recognising apples in different kinds of situations. The proposed model has shown an F1 score of 0.802 which is a significant improvement when compared to already existing detection models.