融合自注意力的SOFC表面缺陷图像分割  

Image Segmentation of SOFC Surface Defects with Fused Self-attention

作　　者：汪尧坤 付晓薇[1,2] 李曦[3] 徐威 WANG Yao-Kun;FU Xiao-Wei;LI Xi;XU Wei(School of Computer Science and Technology,Wuhan University of Science and Technology,Wuhan 430065,China;Wuhan University of Science and Technology Hubei Province Key Laboratory of Intelligent Information Processing and Real-time Industrial System,Wuhan 430065,China;School of Artificial Intelligence and Automation,Huazhong University of Science and Technology,Wuhan 430074,China)

机构地区：[1]武汉科技大学计算机科学与技术学院,武汉430065 [2]武汉科技大学智能信息处理与实时工业系统湖北省重点实验室,武汉430065 [3]华中科技大学人工智能与自动化学院,武汉430074

出　　处：《计算机系统应用》2024年第8期108-114,共7页Computer Systems & Applications

基　　金：国家自然科学基金(61873323,U2066202);深圳科技创新基础研究重点项目(JCYJ20210324115606017);校企合作横向课题(DH1100018);华中科技大学材料成形与模具技术国家重点实验室开放课题(K50059)。

摘　　要：固体氧化物燃料电池(SOFC)表面缺陷的图像分割,对单片SOFC质量检测具有重要意义.针对单片SOFC表面缺陷图像边缘模糊、背景复杂等问题,提出一种融合自注意力的SOFC表面缺陷图像分割方法.首先,提出多通道自注意力模块,以增强多通道间关联和提升通道表示;其次,利用多尺度注意力融合模块,进一步提升网络对不同尺度缺陷特征的提取能力;最后,提出三元联合损失函数对训练过程进行监督.实验表明,提出方法在提升网络分割性能的同时可有效提取单片SOFC表面缺陷.The image segmentation of surface defects on solid oxide fuel cell(SOFC)is of great significance for the quality inspection of monolithic SOFC.Aiming at the problems of blurred edges and complex backgrounds of surface defect images of monolithic SOFC,this study proposed a self-attention fusion method for SOFC surface defect image segmentation.Firstly,a multi-channel self-attention module is proposed to enhance the inter-channel correlation and improve the channel representation.Secondly,a multi-scale attention fusion module is utilized to further improve the network’s ability to extract defect features at different scales;and finally,a triplet joint loss function is proposed to supervise the training process.Experiments show that the proposed method can effectively extract surface defects of monolithic SOFC while improving network segmentation performance.

关 键 词：缺陷分割 固体氧化物燃料电池 损失函数 自注意力 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TM911.4[自动化与计算机技术—计算机科学与技术]