基于深度卷积神经网络的图像语义分割

摘要

近年来, 以深度卷积神经网络 (DEEP Convolutional Neural Network, DCNN) 为代表结合条件随机场 (Conditional Random Field, CRF) 的深度学习算法在图像分割领域中有非常出色的表现.文中首先介绍传统的深度卷积神经网络在图像分割中面临的三个问题以及所借鉴的相关理论;其次介绍对传统深度卷积神经网络三个方面的改进;最后是本次实验的结果和分析.%In recent years, combined Conditional Random Field (CRF), the deep learning algorithm represented by DEEP Convolutional Neural Network (DCNN) has outstanding performance in the image segmentation field. Firstly, this paper introduces three problems faced by traditional DCNN in image segmentation and the relevant theories referenced in this paper. Secondly, it introduces the improvement of three aspects of traditional DCNN. Finally, it gives experimental results and analysis.