PYS: A classification and extraction model of photovoltaics for providing more detailed data to support photovoltaic sustainable development

Photovoltaic power stations have grown unprecedentedly over the last few years. Accurate information regarding the photovoltaic (PV) types and spatial information is crucial for estimating power generation, modeling the environmental impact, and policymaking. However, most previous studies only focused on how to further improve the accuracy of PV spatial information extraction ignoring the importance of distinguishing PV types. Here we propose an instance segmentation model, named PYS, which can directly obtain PV type and spatial information. The PYS is an improved model that incorporates LSPPODC module, BiFormer module, and Wise-IoUv3 loss function into the YOLOv8l-seg model, thereby enhancing the PV extraction and classification performance. Experiments reveal that the PYS model effectively extracts and classifies photovoltaics into four types: land, roof, floating water, and stationary water photovoltaics. The Mask-mIoU, Mask-mAP and Box-mAP of the PYS model achieved 83.2 %, 92.5 % and 92.6 % respectively, outperforming other advanced instance segmentation models in terms of PV extraction and classification. The PYS model can help improve the accuracy and pertinence of extracting and classifying photovoltaics from remote sensing images, as well as provide critical information for the sustainable development of solar energy.