| Title |
Data augmentation with generative adversarial network for solar panel segmentation from remote sensing imagery / |
| Authors |
Lekavičius, Justinas ; Gružauskas, Valentas |
| DOI |
10.5281/zenodo.11522463 |
| Full Text |
|
| Is Part of |
International conference of environmental remote sensing and GIS, Zagreb, 2024.. Zagreb : University of Zagreb. 2024, p. 187-190.. ISSN 3043-8764 |
| Keywords [eng] |
solar panels ; semantic segmentation ; data augmentation ; generative adversarial network ; remote sensing |
| Abstract [eng] |
With the increasing popularity of solar energy in the electricity market, demand arises for data such as precise locations of solar panels for efficient energy planning and management. However, this data is not easily accessible; information such as precise locations sometimes does not exist. Furthermore, existing data sets for training semantic segmentation models of photovoltaic installations are limited, and the manual annotation of remote sensing imagery is time-consuming and labour intensive. Therefore, the pix2pix generative adversarial network (GAN) is used to create additional remote sensing data, enriching the original resampled training data of varying ground sampling distances without compromising its integrity. Experiments with the DeepLabV3 model, ResNet-50 backbone, and pix2pix generative adversarial network architecture were conducted to discover the advantage of using GAN-based data augmentations for a more accurate remote sensing imagery segmentation model. The result is a fine-tuned solar panel semantic segmentation model, trained using transfer learning and an optimal amount – 60% of GAN-generated RS imagery for additional training data. The findings demonstrate the benefits of using GAN-generated images as additional training data, addressing the issue of limited data sets, and increasing IoU and F1 metrics by 2% and 1.46%, respectively, compared to classic augmentations. The improved semantic segmentation model allows for better solar panel detection in remote sensing images and the potential development of a regional photovoltaic installation map for better electricity network planning and risk management. |
| Published |
Zagreb : University of Zagreb |
| Type |
Conference paper |
| Language |
English |
| Publication date |
2024 |
| CC license |
|
