Identifying solar cell efficiency shortfalls in production lines is crucial to troubleshoot and optimize manufacturing processes. With the adoption of luminescence imaging as a key end-of-line characterization tool, a wealth of information is available to evaluate cell performance and classify defects, suitable for user input-free deep-learning analysis. We propose an automated reconstruction method, based on state-of-the-art generative adversarial networks, to remove defective regions in luminescence images. The reconstructed and original images are compared to estimate the efficiency loss. The method is validated on intentionally damaged cells by reconstructing defect-free images and then predicting the efficiency loss. The method can differentiate between different types of defects and pinpoint the defects that lead to the highest efficiency shortfall, enabling manufacturers to troubleshoot production lines in a fast and cost-effective manner. The proposed approach unlocks the potential of luminescence imaging as an effective end-of-line characterization tool.

识别生产线中的太阳能电池效率不足对于排除故障和优化制造工艺至关重要。随着采用发光成像作为关键的终端表征工具，大量信息可用于评估细胞性能和分类缺陷，适用于用户无需输入的深度学习分析。我们提出了一种基于最先进的生成对抗网络的自动重建方法，以去除发光图像中的缺陷区域。将重建图像和原始图像进行比较以估计效率损失。通过重建无缺陷图像然后预测效率损失，该方法在故意损坏的细胞上得到验证。该方法可以区分不同类型的缺陷并查明导致最高效率不足的缺陷，使制造商能够以快速且具有成本效益的方式对生产线进行故障排除。所提出的方法释放了发光成像作为有效的终端表征工具的潜力。