L-FCN: A lightweight fully convolutional network for biomedical semantic segmentation
Kaiyue Li, Guangtai Ding, Haitao Wang · 2018
For the past few years, deep learning-based methods have been widely used in the field of biomedical imaging. In biomedical image processing, the typical application of deep learning is semantic segmentation. However, the classical deep learning methods require higher hardware consumption and computational costs. In order to resolve this problem, we propose a new lightweight fully convolutional network (L-FCN). L-FCN consists of traditional watershed algorithm and fully convolutional network, which eliminates useless non-edge pixels to improves the efficiency of semantic segmentation. Experimental evaluations on ISBI 2012 dataset indicate that our L-FCN algorithm outperforms U-net in time efficiency and computational costs, meanwhile, maintain approximate image segmentation effect. Due to its portability, L-FCN can be applied to many biomedical areas easily.