EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Advanced Defect Detection in Photovoltaic Modules: Integrating Attention Mechanisms and Depthwise Separable Convolutions in YOLOv7

Yu Tian, Lanyang Zhang, Gaoming Wang, Zhiqiang Liu, Shaofeng Wu

Data and Metadata · 12.03.2026 · 5 · 1414

DOI: 10.56294/dm20261414 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Data and Metadata, online erschienen am 12.03.2026. Verfasst von Yu Tian und 4 weiteren Autor:innen. Die Arbeit nennt 15 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Defect detection of photovoltaic (PV) modules is crucial for ensuring the power generation efficiency and operational safety of PV power stations. To address the challenge of balancing detection speed and accuracy in complex backgrounds using existing object detection algorithms, this paper proposes an improved defect detection algorithm based on YOLOv7.this study proposes an enhanced defect detection algorithm based on an improved YOLOv7 framework. First, a lightweight and efficient FasterNet is adopted as the backbone feature extraction module to optimize feature representation while reducing computational overhead. Second, Depth wise Separable Convolution modules are introduced into the detection head to replace part of the standard convolutions, which significantly re-duces the number of model parameters and computational complexity. Furthermore, to compensate for potential accuracy loss due to network lightweighting, an attention mechanism combined with deformable convolution is integrated into the detection head, thereby enhancing the model's ability to recognize irregular defects and improving localization accuracy. Experimental results demonstrate that the proposed algorithm achieves high detection accuracy while maintaining efficient inference speed. On a PV module defect dataset, the mean average precision (mAP@0.5) reaches 83.3%, representing an improvement of 1.4% over the baseline YOLOv7. The parameter count is reduced by 4.6 M, computational load is decreased by approximately 60%, and inference speed is accelerated to 1.8ms per frame. The proposed method can rapidly and accurately identify and localize three typical types of PV module defects—cracks, grid defects, and spots—under complex real-world scenarios, providing an effective technical solution for intelligent inspection during PV module production and operational maintenance.

Zitieren

APA

Tian, Y., Zhang, L., Wang, G., Liu, Z., & Wu, S. (2026). Advanced Defect Detection in Photovoltaic Modules: Integrating Attention Mechanisms and Depthwise Separable Convolutions in YOLOv7. Data and Metadata, 5, 1414. https://doi.org/10.56294/dm20261414

Vancouver

Tian Y, Zhang L, Wang G, Liu Z, Wu S. Advanced Defect Detection in Photovoltaic Modules: Integrating Attention Mechanisms and Depthwise Separable Convolutions in YOLOv7. Data and Metadata. 2026;5:1414. doi:10.56294/dm20261414

BibTeX
@article{Tian2026,
  author = {Tian, Yu and Zhang, Lanyang and Wang, Gaoming and Liu, Zhiqiang and Wu, Shaofeng},
  title = {Advanced Defect Detection in Photovoltaic Modules: Integrating Attention Mechanisms and Depthwise Separable Convolutions in YOLOv7},
  journal = {Data and Metadata},
  year = {2026},
  volume = {5},
  pages = {1414},
  doi = {10.56294/dm20261414}
}
Kostenlos starten →