深度神经网络图像语义分割方法综述

Survey of Image Semantic Segmentation Methods Based on Deep Neural Network

图像语义分割是计算机视觉领域近年来的热点研究课题,随着深度学习技术的兴起,图像语义分割与深度学习技术进行融合发展,取得了显著的进步,在无人驾驶、智能安防、智能机器人、人机交互等真实场景中应用广泛。首先对应用于图像语义分割的几种深度神经网络模型进行简单介绍,接着详细阐述了现有主流的基于深度神经网络的图像语义分割方法,依据实现技术的区别对图像语义分割方法进行分类,并对每类方法中代表性算法的技术特点、优势和不足进行分析与总结。之后归纳了图像语义分割常用的大规模公共数据集和性能评价指标,并在此基础上对经典的语义分割方法的实验结果进行了对比,最后对语义分割领域未来可行的研究方向进行展望。

Image semantic segmentation is a hot research topic in the field of computer vision in recent years.With the rise of deep learning technology,image semantic segmentation and deep learning technology are integrated and developed,which has made significant progress.It is widely used in practical scenarios such as unmanned driving,intelligent security,intelligent robot,human-computer interaction.Firstly,several deep neural network models for image semantic segmentation are introduced,and then the existing mainstream deep neural network-based image semantic segmentation methods are introduced.According to the differences of implementation technologies,image semantic segmentation methods are classified,and the technical characteristics,advantages and disadvantages of representative algorithms are analyzed and summarized.After that,the common datasets and performance evaluation indexes of image semantic segmentation are summarized,and the experimental results of classic semantic segmentation methods are compared on this basis.Finally,the future feasible research directions in the field of semantic segmentation are prospected.