基于归一化控制器的光伏图像无监督域适应缺陷检测

陈海永; 史世杰

doi:10.19912/j.0254-0096.tynxb.2023-0418

PDF(2981 KB)

太阳能学报 ›› 2024, Vol. 45 ›› Issue (7) : 540-547. DOI: 10.19912/j.0254-0096.tynxb.2023-0418

基于归一化控制器的光伏图像无监督域适应缺陷检测

陈海永, 史世杰

作者信息 +

NORMALIZED CONTROLLER FOR PHOTOVOLTAIC DEFECT DETECTION IN DYNAMIC OPEN SCENARIOS

Chen Haiyong, Shi Shijie

Author information +

文章历史 +

摘要

为解决在太阳电池数据集域偏移问题,提出一种针对动态开放场景下光伏缺陷检测的数据归一化控制器（DNC）,以提高网络的域适应能力。在测试阶段,该文的DNC方法通过修改模型参数,根据小批次样本数据（0.5%）来实现领域统计纠正。DNC可将目标域产生域偏移的数据映射到与源域数据相同的分布空间,而无需提前标注数据或访问目标域的全部数据。实验结果表明,DNC能显著提高目标检测模型对域偏移数据的适应能力。仅使用少量未标记的目标域数据（少于0.5%）就可在分布外数据上获得显著的性能提升,同时不会降低模型的预测速度（FPS）。

Abstract

To address domain shift issues in solar cell datasets, this thesis proposes a data normalization controller （DNC） tailored for photovoltaic defect detection in dynamic open environments, aiming to enhance the network's domain adaptation capability. During the testing phase, the DNC method in this study adjusts model parameters based on small batches of sample data （less than 0.5%）, effectively rectifying domain statistics. DNC maps data experiencing domain shift in the target domain onto the same distribution space as the source domain data, without requiring prior labeling of data or access to the entire target domain dataset. Experimental results demonstrate that DNC significantly improves the target detection model's adaptability to domain-shifted data. Using only a minimal amount of unlabeled target domain data （less than 0.5%）, substantial performance gains can be achieved on out-of-distribution data, while maintaining the model's prediction speed （FPS）.

导出引用

陈海永, 史世杰. 基于归一化控制器的光伏图像无监督域适应缺陷检测[J]. 太阳能学报. 2024, 45(7): 540-547 https://doi.org/10.19912/j.0254-0096.tynxb.2023-0418

Chen Haiyong, Shi Shijie. NORMALIZED CONTROLLER FOR PHOTOVOLTAIC DEFECT DETECTION IN DYNAMIC OPEN SCENARIOS[J]. Acta Energiae Solaris Sinica. 2024, 45(7): 540-547 https://doi.org/10.19912/j.0254-0096.tynxb.2023-0418

中图分类号： TK391.4

参考文献

[1] DEITSCH S, CHRISTLEIN V, BERGER S, et al.Automatic classification of defective photovoltaic module cells in electroluminescence images[J]. Solar energy, 2019, 185: 455-468.
[2] 周颖, 毛立, 张燕, 等. 改进CNN的太阳电池缺陷识别方法研究[J]. 太阳能学报, 2020, 41(12): 69-76.
ZHOU Y, MAO L, ZHANG Y, et al.Research on defect detection and classification for solar cells based on improved convolutional neural network[J]. Acta energiae solaris sinica, 2020, 41(12): 69-76.
[3] 周颖, 叶红, 王彤, 等. 基于多尺度 CNN 的光伏组件缺陷识别[J]. 太阳能学报, 2022, 43(2): 211-216.
ZHOU Y, YE H, WANG T, et al.Photovoltaic modules defect identification base on multi-scale convolution neural network[J]. Acta energiae solaris sinica, 2022, 43(2): 211-216.
[4] SU B Y, CHEN H Y, CHEN P, et al.Deep learning-based solar-cell manufacturing defect detection with complementary attention network[J]. IEEE transactions on industrial informatics, 2021, 17(6): 4084-4095.
[5] CHEN C, FU Z H, CHEN Z H, et al.HoMM: higher-order moment matching for unsupervised domain adaptation[C]//AAAI Conference on Artificial Intelligence. New York, USA, 2020: 3422-3429.
[6] LUO Y W, ZHENG L, GUAN T, et al.Taking a closer look at domain shift:category-level adversaries for semantics consistent domain adaptation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach, CA, USA, 2019: 2502-2511.
[7] LIN T Y, GOYAL P, GIRSHICK R, et al.Focal loss for dense object detection[C]//2017 IEEE International Conference on Computer Vision (ICCV). Venice, Italy, 2017: 2999-3007.
[8] RECHT B, ROELOFS R, SCHMIDT L, et al. Do ImageNet classifiers generalize to ImageNet? [EB/OL].2019: arXiv: 1902.10811. http://arxiv.org/abs/1902.10811.
[9] HE Z W, ZHANG L.Multi-adversarial faster-RCNN for unrestricted object detection[C]//2019 IEEE/CVF International Conference on Computer Vision (ICCV). Seoul, Korea, 2019: 6667-6676.
[10] ZHENG Y T, HUANG D, LIU S T, et al.Cross-domain object detection through coarse-to-fine feature adaptation[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Seattle, WA, USA, 2020: 13763-13772.
[11] WANG H Y, ZHU Y K, ADAM H, et al.MaX-DeepLab: end-to-end panoptic segmentation with mask transformers[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Nashville, TN, USA, 2021: 5459-5470.
[12] WANG Y, ZHANG R, ZHANG S, et al.Domain-specific suppression for adaptive object detection[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Nashville, TN, USA, 2021: 9598-9607.
[13] SU P, WANG K, ZENG X Y, et al.Adapting object detectors with conditional domain normalization[C]//Computer Vision-ECCV 2020: 16th European Conference. Glasgow, UK, 2020: 403-419.
[14] XU C D, ZHAO X R, JIN X, et al.Exploring categorical regularization for domain adaptive object detection[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Seattle, WA, USA, 2020: 11721-11730.
[15] CHEN C Q, ZHENG Z B, DING X H, et al.Harmonizing transferability and discriminability for adapting object detectors[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Seattle, WA, USA, 2020: 8866-8875.
[16] XU M H, WANG H, NI B B, et al.Cross-domain detection via graph-induced prototype alignment[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Seattle, WA, USA, 2020: 12352-12361.
[17] ZHU J Y, PARK T, ISOLA P, et al.Unpaired image-to-image translation using cycle-consistent adversarial networks[C]//2017 IEEE International Conference on Computer Vision (ICCV). Venice, Italy, 2017: 2242-2251.
[18] 杨振宇, 苏建徽, 汪海宁, 等. 光伏阵列I-V特性曲线测量误差分析与修正方法[J]. 太阳能学报, 2023, 44(5): 210-216.
YANG Z Y, SU J H, WANG H N, et al.Measurement error analysis and correction for I-V characteristic of photovoltaic array[J]. Acta energiae solaris sinica, 2023, 44(5): 210-216.
[19] 张东, 俞凯, 景金龙, 等. 积尘形态及密度对太阳能PV/T系统的性能影响研究[J]. 太阳能学报, 2023, 44(5): 171-177.
ZHANG D, YU K, JING J L, et al.Effect of dust morphology and density on performance of solar PV/T system[J]. Acta energiae solaris sinica, 2023, 44(5): 171-177.
[20] 王道累, 李明山, 姚勇, 等. 改进SSD的光伏组件热斑缺陷检测方法[J]. 太阳能学报, 2023, 44(4): 420-425.
WANG D L, LI M S, YAO Y, et al.Method of hotspot detection of photovoltaic panels modules on improved SSD[J]. Acta energiae solaris sinica, 2023, 44(4): 420-425.