Development assessment of regional rooftop photovoltaics based on remote sensing and deep learning

Assessing the development of rooftop photovoltaic (PV) plays a positive role in promoting the deployment of solar installations. In response to the problem that previous studies did not consider the PV already installed on rooftops and thus had a low level of refinement, this study proposes a dual-branch framework based on remote sensing imagery and deep learning to effectively monitor the current status and estimate the future potential of regional rooftop PV installations. Based on the dual-branch framework, a semantic segmentation network called MANet is designed that integrated multi-attention modules for more accurate extraction of roofs and PV panels from remote sensing imagery. Finally, the methodology is validated through an application in Taihuyuan Town, Hangzhou City. Model comparisons show that MANet achieves the best accuracy in both roof and PV panel extraction, with IoU scores of 88.17% and 91.58%, respectively. The difference between the existing PV installed capacity and the roof area extracted by the framework and the corresponding statistical data is 8.1% and 11.7%, respectively. The framework calculates the existing rooftop PV installed capacity to be 16,173.74 kW, and estimates the PV installed capacity potential of the remaining rooftops to be 403 MW. •A 2-branch framework for rooftop PV monitoring and potential assessment was designed.•Higher accuracy semantic segmentation algorithms for roofs and PV panels was built.•1.08×105m2 existed rooftop PV with 16.17 MW capacity were infered for Taihuyuan.•403 MW of rooftop PV potential was estimated for Taihuyuan Town in 2023.