摘要
针对光伏用电致发光缺陷检测仪空间分辨率目视判定重复性、准确性差,并且难以量化等问题,提出一套基于卷积神经网络模型的空间分辨率量化评估方法。参照相关标准JJF(闽)1088-2018,采取不同的拍摄条件拍摄并切割出空间分辨率线对图像,对图像进行人工分类。设计卷积神经网络结构,采用卷积层、池化层和全连接层结构,并使用已分类完成的空间分辨率线对图像对模型进行训练。最终使用测试集对模型进行评估,结果表明,模型在测试集上的判别正确率达到99.2%。该方法满足使用要求,可取代目视判别,提高了光伏用电致发光缺陷检测仪空间分辨率判定的准确性与重复性,并为其量化提出了新的解决思路。由于太阳电池片用光致发光缺陷检测仪与光伏用电致发光缺陷检测仪的成像方式类似,该方法可兼容太阳电池片光致发光检测仪的性能评估。
To overcome poor accuracy, quantification difficulty, and repeatability of visual judgment of the spatial resolution of a photovoltaic electroluminescent defect detector, different image capturing conditions were adopted based on standards JJF(Min)1088-2018. Further, the images of the spatial resolution line pairs were cut and classified. The convolution neural network employed included a convolution layer, a pooling layer, and a full connection layer. The classified spatial resolution line pairs were used to train the model and were evaluated with test samples. The results show that the discriminant accuracy of the model reaches 99.2% for the test samples. This method satisfies the application requirements and can replace visual discrimination that provides a new solution for quantifying the spatial resolution of photovoltaic electroluminescent defect detector. Given that photoluminescence defect detectors for solar cells have similar imaging mode to electroluminescence defect detectors, this method can be compatible with the performance evaluation of the photoluminescence defect detectors for solar cells.
作者
何翔
杨爱军
黎健生
罗海燕
陈彩云
HE Xiang;YANG Ai-jun;LI Jian-sheng;LUO Hai-yan;CHEN Cai-yun(Fujian Institute of Metrology,Fuzhou 350003,China)
出处
《光学精密工程》
EI
CAS
CSCD
北大核心
2020年第3期542-547,共6页
Optics and Precision Engineering
基金
福建省市场监督管理局科技资助项目(No.FJQI2018014)。
关键词
光伏
电致发光
缺陷检测仪
空间分辨率
卷积神经网络
光致发光
photovoltaic
electroluminescent
defect detection
spatial resolution
convolutional neural network
photoluminescence
作者简介
何翔(1990-),男,福建福州人,助理工程师,2013年、2016年于中南大学分别获得学士、硕士学位,主要从事从光伏计量方面的研究。E-mail:315107763@qq.com;杨爱军(1973-),女,福建福州人,高级工程师,主要从事光伏计量和电学计量的研究。E-mail:yaj@fjjl.net。
