Photovoltaic Module Detection Models and Training Datasets

This training dataset targets photovoltaic (PV) module segmentation. The data was used to train U-Net segmentation models and YOLOv8 instance segmentation models, which are available to the public. The dataset contains images annotated with over 3000 PV module segmentation masks in 136 RGB and 74 thermal images. The segmentation labels are provided both as segmentation mask images suitable for U-Net training and as text files with labeled contours suitable for YOLO training. The dataset was annotated using the Grid Annotation Tool. It contains metadata from the annotation process, and allows the user to alter the original annotations. An additionall dataset for full PV installation segmentation, which does not differentiate between individual PV modules, is also included.

When using this dataset, please cite (at least one of) the related publications.

Annotated PV module segmentation dataset (700MB)

UNET PV module and PV installation segmentation models (344MB)

YOLOv8n PV module segmentation models (34MB)

• Kozák, V., Košnar, K., Chudoba, J., Kulich M., and Přeučil, L. (2025). Visual Localization via Semantic Structures in Autonomous Photovoltaic Power Plant Inspection. Arxiv preprint.
  - The dataset originates from this publication.
  URL  BibTeX

  @misc{kozak2025PVLocalization,
    author = {Viktor Kozák and Karel Košnar and Jan Chudoba and Miroslav Kulich and Libor Přeučil},
    title = {Visual Localization via Semantic Structures in Autonomous Photovoltaic Power Plant Inspection},
    year = {2025},
    eprint={2501.14587},
    archivePrefix={arXiv},
    primaryClass={cs.CV},
    url = {https://arxiv.org/abs/2501.14587},
    doi = {10.48550/arXiv.2501.14587}
  }
  
  

• Kozák, V., Chudoba, J., and Přeučil, L. (2025). Detailed Aerial Mapping of Photovoltaic Power Plants Through Semantically Significant Keypoints. International Journal of Engineering and Geosciences.
  -This work used and evaluated the segmentation models on images from higher altitudes.
  URL  BibTeX

  @Article{kozak2025MappingPV,
    author = {Kozák, Viktor and Chudoba, Jan and Přeučil, Libor},
    title = {Detailed Aerial Mapping of Photovoltaic Power Plants Through Semantically Significant Keypoints},
    JOURNAL = {International Journal of Engineering and Geosciences},
    VOLUME = {11},
    YEAR = {2025},
    NUMBER = {2},
    PAGES={352–362}, 
    URL = {https://doi.org/10.26833/ijeg.1737764},
    ISSN = {2548-0960},
    DOI = {10.26833/ijeg.1737764}
  }