基于超像素混合合并方法的铜电解槽红外图像功能区域分割  

Segment of functional areas in copper electrolytic cell infrared image based on superpixel hybrid merging method

作　　者：李欣[1] 李勇刚[1] 吴仁超[1] 朱红求[1] 周灿[1] LI Xin;LI Yonggang;WU Renchao;ZHU Hongqiu;ZHOU Can(School of Automation,Central South University,Changsha 410083,China)

机构地区：[1]中南大学自动化学院,湖南长沙410083

出　　处：《中南大学学报（自然科学版）》2023年第1期13-23,共11页Journal of Central South University:Science and Technology

基　　金：国家重点研发计划项目(2019YFB1704700)。

摘　　要：铜电解槽面红外图像由多类功能区域构成,综合各区域温度特征可有效评估电解状态。但各区域灰度分布复杂,区域间边界模糊,导致功能区域难以分割。为了提高功能区域分割的准确性和鲁棒性,提出超像素混合合并方法。首先,针对小尺度空间,使用区域临界图法,并提出基于灰度空间相关性的合并停止条件,避免过拟合。随后,针对大尺度空间,提取超像素纹理特征进行多类别支持向量机(SVM)训练,完成区域分类。研究结果表明:本文所提方法可以有效分割电解槽红外图像中的功能区域,与活动轮廓方法、Otsu方法相比,在区域完整性方面表现更优。Infrared image of copper electrolytic cells surface is composed of multiple functional areas, and electrolysis state can be effectively evaluated by integrating the temperature characteristics of each area. However,gray distribution of each region is complex, and the boundary between regions is fuzzy, which leads to the difficulty of segmenting functional areas. In order to improve the accuracy and robustness of functional region segmentation, a superpixel hybrid merging method was proposed. Firstly, in the small scale space, the region adjacency graph method was used, and a stop condition based on gray spatial correlation was proposed to avoid over merging. Then, for large scale space, superpixel texture features were extracted for multi-category SVM training to complete region classification. The results show that the proposed method can effectively segment functional areas in infrared images, and has better performance in region integrity than active contour method and Otsu method.

关 键 词：功能区域分割 红外图像 超像素 混合合并 支持向量机 

分 类 号：TF355.4[冶金工程—冶金机械及自动化]