计算机技术与发展2026，Vol.36Issue(8)：33-40,8.DOI:10.20165/j.cnki.ISSN1673-629X.2026.0060
基于多尺度特征感知的SOFC表面缺陷检测算法
SOFC Surface Defect Detection Algorithm Based on Multi-scale Feature Perception
摘要
Abstract
The solid oxide fuel cell(SOFC)is a key component of SOFC power generation systems,and its quality directly affects the stable operation of the cell stack and the service life of the system.To address issues such as the random shapes and sizes of surface defects on individual SOFCs,which can lead to false detection and missed detection,a surface defect detection algorithm for SOFCs based on multi-scale feature perception is proposed.Firstly,a multi-scale feature perception(MSFP)module is introduced to capture the positional information of target defects,enhance feature extraction in defect regions,and suppress interference from complex backgrounds.Secondly,an adaptive gated residual attention(AGRA)module is designed to enable the model to simultaneously capture fine-grained local features and global structural information.Finally,the Shape-IoU loss function is employed to focus on the shape and scale of the bounding boxes for surface defects,thereby optimizing the bounding box regression process.Experimental results show that the proposed algorithm achieves mean average precision,precision,and recall rates of 87.5%,85.1%,and 80.2%,respectively,on the SOFC surface defect dataset.Compared with other mainstream object detection algorithms,the proposed algorithm effectively mitigates false detection and missed detection,demonstrates strong robustness.关键词
缺陷检测/固体氧化物燃料电池/特征感知/注意力机制/损失函数Key words
defect detection/solid oxide fuel cell/feature perception/attention mechanism/loss function分类
信息技术与安全科学引用本文复制引用
付晓薇,刘晓,李曦..基于多尺度特征感知的SOFC表面缺陷检测算法[J].计算机技术与发展,2026,36(8):33-40,8.基金项目
国家重点研发计划(2022YFB4002205) （2022YFB4002205）
深圳市基础研究专项自然科学基金(JCYJ20250604191409014) （JCYJ20250604191409014）
校企合作横向课题(DH1101103) （DH1101103）
国家留学基金(202508420230) （202508420230）
