首页|Investigators from Beihang University Release New Data on Robotics (Efficientnet -eca: a Lightweight Network Based On Efficient Channel Attention for Class-imbal anced Welding Defects Classification)
Investigators from Beihang University Release New Data on Robotics (Efficientnet -eca: a Lightweight Network Based On Efficient Channel Attention for Class-imbal anced Welding Defects Classification)
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
NETL
NSTL
2024 OCT 03 (NewsRx)-By a News Reporter-Staff News Editor at Robotics & Machine Learning Daily News Daily News-Fresh data on Robotics are presented i n a new report. According to news reporting from Beijing, People's Republic of C hina, by NewsRx journalists, research stated, "Welding defects recognition is cr ucial for ensuring weldment quality in robot arc welding. Nevertheless, due to u nbalanced data distributions, limited computing resources in factory, as well as intraclass variability and interclass similarity among different welding defect s, it is difficult to extract the most discriminative defect features from weldi ng molten pool images on site, resulting in weak class-imbalanced defect recogni tion performance." The news correspondents obtained a quote from the research from Beihang Universi ty, "To address the above issue, a novel lightweight network named EfficientNet- ECA is proposed for recognizing and classifying welding defects according to mol ten pool images of robot arc welding. Firstly, the EfficientNet- ECA network base d on Efficient Channel Attention (ECA) is designed to extract the most discrimin ative features of different defects from molten pool images, where ECA is employ ed to enhance cross-channel information interactions. Secondly, a dynamic equali zed focal loss function is proposed to both rebalance the loss contribution of c lass-imbalanced data of different defects and improve defects recognition accura cy. Subsequently, a class-imbalanced dataset containing eight typical welding de fects is constructed to evaluate the EfficientNet-ECA model. Finally, the propos ed model is comprehensively analyzed through ablation studies and compared with the existing state-of-the-art lightweight models, and results show that the prop osed method exhibits better effectiveness and generalizability in classifying cl ass-imbalanced defects across different welding scenarios, achieving the highest accuracy of 95.84% on a self-constructed AL5083 dataset and 96.50 % on a publicly available SS304 dataset."
BeijingPeople's Republic of ChinaAsi aEmerging Technologies,Machine LearningRobotRoboticsBeihang University
NETL
NSTL
