elpv-dataset

Dataset Overview

Dataset Name

• A Benchmark for Visual Identification of Defective Solar Cells in Electroluminescence Imagery

Dataset Content

• Contains 2,624 samples, each a 300×300‑pixel 8‑bit grayscale image, showing functional and defective solar cells with varying defect severity.
• Images originate from 44 different solar modules, with defect types including intrinsic and extrinsic.

Dataset Characteristics

• All images are standardized in size and orientation.
• Any distortion caused by camera lenses has been removed prior to solar cell extraction.

Dataset Annotation

• Each image is accompanied by a defect probability (floating‑point value between 0 and 1) and solar module type (monocrystalline or polycrystalline).
• Images are stored in the images directory, with corresponding annotations in the labels.csv file.

Dataset Usage

• Use the utils/elpv_reader module in Python to load images and annotations.
• Example code:

from elpv_reader import load_dataset
images, proba, types = load_dataset()

• Requires NumPy and Pillow libraries.

Dataset Citation

• When using this dataset for scientific research, cite the following publications:
  • Buerhop-Lutz, C. et al. (2018). A Benchmark for Visual Identification of Defective Solar Cells in Electroluminescence Imagery.
  • Deitsch, S. et al. (2021). Segmentation of photovoltaic module cells in uncalibrated electroluminescence images.
  • Deitsch, S. et al. (2019). Automatic classification of defective photovoltaic module cells in electroluminescence images.

Dataset License

• The dataset is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
• Commercial use requires contacting the dataset provider for further information.