基于DCGANs的半片光伏组件电致发光图像增强技术

针对半片光伏组件电致发光（electroluminescence,EL）缺陷自动识别过程中训练用样本不足导致模型过拟合的问题,采用深度卷积生成对抗网络（deep convolutional generative adversarial networks,DCGANs）生成可控制属性的半片光伏组件EL图像，再采用多尺度结构相似性（multiscale structural similarity,MS-SSIM）指标对生成的EL图像与拍摄的EL图像之间的相似程度进行了评估。评估结果得到，使用DCGANs生成的所有类型半片光伏组件的EL图像与拍摄的EL图像的MS-SSIM指标都大于0.55，大部分的MS-SSIM值在0.7附近。在分类模型的训练过程中，测试集准确率随着训练集中生成图像数量的增加而升高，当生成图像数量达到6 000张时，测试集准确率达到97.92%。实验结果表明，采用DCGANs能够生成高质量且可控制属性的半片光伏组件EL图像，较好地解决因缺少训练样本而导致的模型过拟合问题。

Electroluminescence image enhancement technology of half-cut photovoltaic module based on DCGANs

Aiming at the problem of model overfitting caused by insufficient training samples in the automatic identification process of electroluminescence (EL) defects for half-cut photovoltaic modules, the deep convolutional generative adversarial networks (DCGANs) were adopted to generate the half-cut photovoltaic module EL images with controllable attributes. The similarity between the generated EL images and the captured EL images was evaluated by using the multi-scale structural similarity (MS-SSIM) index. The evaluation results show that the MS-SSIM indexes of all types of EL images generated by DCGANs and captured EL images are greater than 0.55, and most of the MS-SSIM values are near 0.7. In the training process of the classification models, the accuracy of the test set increases with the increase of the number of images generated in the training set. When the number of generated images reaches 6 000, the accuracy of the test set reaches 97.92%. The experimental results show that the DCGANs can generate half-cut photovoltaic module EL images with high quality and controllable attributes, which can better solve the problem of model overfitting caused by the lack of training samples.