Multi-resolution dataset for photovoltaic panel segmentation from satellite and aerial imagery

[摘要] In the context of global carbon emission reduction, solarphotovoltaic (PV) technology is experiencing rapid development. Accurate localized PVinformation, including location and size, is the basis for PV regulation andpotential assessment of the energy sector. Automatic information extractionbased on deep learning requires high-quality labeled samples that should becollected at multiple spatial resolutions and under different backgroundsdue to the diversity and variable scale of PVs. We established a PV datasetusing satellite and aerial images with spatial resolutions of 0.8, 0.3, and0.1 m, which focus on concentrated PVs, distributed ground PVs, andfine-grained rooftop PVs, respectively. The dataset contains 3716 samples ofPVs installed on shrub land, grassland, cropland, saline–alkali land, and water surfaces, as well as flat concrete, steel tile, and brick roofs. The datasetis used to examine the model performance of different deep networks on PVsegmentation. On average, an intersection over union (IoU) greater than85 % is achieved. In addition, our experiments show that direct crossapplication between samples with different resolutions is not feasible andthat fine-tuning of the pre-trained deep networks using target samples isnecessary. The dataset can support more work on PV technology for greater value, suchas developing a PV detection algorithm, simulating PV conversion efficiency,and estimating regional PV potential. The dataset is available from Zenodoon the following website: https://doi.org/10.5281/zenodo.5171712 (Jiang et al., 2021).

[学科分类] 眼科学