基于深度学习的图像实例分割技术研究进展  

Research on the Progress of Image Instance Segmentation Based on Deep Learning

作　　者：梁新宇 林洗坤 权冀川 肖铠鸿 LIANG Xin-yu;LIN Xi-kun;QUAN Ji-chuan;XIAO Kai-hong(College of Command & Control Engineering,Army Engineering University of PLA,Nanjing,Jiangsu 210007,China;College of Software Engineering,Huazhong University of Science & Technology,Wuhan,Hubei 430070,China;Unit 73676 of PLA,Wuxi,Jiangsu 214400)

机构地区：[1]陆军工程大学指挥控制工程学院,江苏南京210007 [2]华中科技大学软件学院,湖北武汉430070 [3]中国人民解放军73676部队,江苏无锡214400

出　　处：《电子学报》2020年第12期2476-2486,共11页Acta Electronica Sinica

摘　　要：随着深度学习算法在图像分割领域的成功应用,在图像实例分割方向上涌现出一大批优秀的算法架构.这些架构在分割效果、运行速度等方面都超越了传统方法.本文围绕图像实例分割技术的最新研究进展,对现阶段经典网络架构和前沿网络架构进行梳理总结,结合常用数据集和权威评价指标对各个架构的分割效果进行比较和分析.最后,对目前图像实例分割技术面临的挑战以及可能的发展趋势进行了展望.With the successful application of deep learning algorithms in the field of image segmentation,a large number of excellent algorithm architectures have emerged in the direction of image instance segmentation.These architectures surpass the traditional methods in terms of segmentation effects and running speed.This paper focuses on the latest research progress of image instance segmentation technology,summarizes the current classic network architecture and cutting-edge network architecture,and uses common datasets and authoritative evaluation indicators to compare and analyze the segmentation effects of each architecture.Finally,the challenges and possible development trends of image instance segmentation technology are prospected.

关 键 词：深度学习 图像分割 实例分割 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]