粘连颗粒图像的分割方法综述  

A Review of Segmentation Methods for Adhesive Particle Images

作　　者：高辉[1] 甄彤[1] 李智慧[1] Gao Hui;Zhen Tong;Li Zhihui(Key Laboratory of Grain Information Processing and Control,College of Information Science and Engineering,Henan University of Technology,Zhengzhou 450001)

机构地区：[1]粮食信息处理与控制重点实验室,河南工业大学信息科学与工程学院,郑州450001

出　　处：《中国粮油学报》2022年第3期186-194,共9页Journal of the Chinese Cereals and Oils Association

基　　金：国家科技支撑计划(2018YFD0401404)。

摘　　要：粘连图像分割作为颗粒计数、分类、定级评价、识别的基础环节,其实际应用价值不言而喻。本文简要介绍现有的传统分割算法和基于深度学习的分割算法种类,根据粘连颗粒尺寸小、随机散落、数量众多、形状不规则及边缘特征模糊等特点,结合粘连分割算法在各种领域中的应用现状,重点阐述基于分水岭、凹点、U-Net语义分割的方法,介绍关键技术,分析优缺点,明确适用范围,进行算法评价,并对相关的研究方向和发展趋势作出展望。As a basic link of particle counting, classification, recognition, grading evaluation, adhesion image segmentation had self-evident practical application value. The present paper gave a brief introduction of the existing traditional segmentation algorithms and the kinds of segmentation algorithm based on deep learning. According to small adhesive particle size small, random scattering, large quantity, irregular shape, fuzzy edge character and other features, and combining with the application status of adhesion and partitioning algorithm in various fields, the present paper emphasized on the methods based on watershed, concave point and U-Net semanteme division, introduced key technologies, analyzed the advantages and disadvantages, cleared the applicable scape, conducted algorithm application and prospected the direction and development trend of relevant research.

关 键 词：粘连分割 分水岭 凹点 U-Net语义分割 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]