基于深度学习的SOFC球壳结构检测方法

为优化固体氧化物燃料电池(SOFC)制备工艺提供数据支持和理论依据,提出一种基于光学显微镜微观图像的球壳检测方法。在SOFC阳极微观图像上如果出现球壳结构,表明氧化镍(NiO)未完全还原,该现象严重影响电池的电化学性能、稳定性和使用寿命。为此,利用深度学习方法对SOFC光学显微镜图像进行球壳结构检测,分析阳极NiO的还原程度,通过预选框尺度、网络结构及参数的优化来提高检测性能。为充分利用有限的数据训练网络模型,对训练数据进行扩增。实验结果表明,该检测方法可准确有效地检测与识别形状复杂的SOFC阳极球壳结构,具有检测速度快,球壳结构定位精度较高等优点。

SOFC core-shell structure detection method based on deep learning

To provide data support and theory basis for optimizing the preparation process of solid oxide fuel cell(SOFC),a core-shell detection method based on optical microscopic(OM)images was proposed.When the core-shell structure appears in images,it indicates that NiO is not completely reduced,which seriously affects the electrochemical performance,stability and service life of the SOFC.To this end,a method based on deep learning was proposed for detecting the core-shell of OM image and analyzing the reduction of NiO.The detection performance was improved by pre-selecting frame scale,network structure and parameter optimization.To make full use of the limited data,the training data were amplified.Experimental results show that the proposed method can accurately and effectively recognize the core-shell structure with complex shape in the SOFC image.It achieves higher detection speed and positioning accuracy of core-shell.