Deep convolution and up-convolution network for plant segmentation

Abstract

In this study, we propose a deep learning method to segment plants in images. The deep learning method is composed of a contracting path and expanding path. The contracting path learns high level feature representation of images and the expanding path interprets the high level features and generates segmentation maps. The proposed method is trained and validated, via five-fold cross validation, using images of radish seedlings. The method achieved 99.15% accuracy and 0.9790 Dice coefficient, suggesting that deep learning could play a significant role in processing and analyzing plant images.