Photovoltaic Panel Identification and Potential Assessment Based on Remote Sensing and Deep Learning: A Case Study of Yancheng City

The rapid expansion of PV power stations necessitates precise spatial mapping and balanced development planning, yet trade-offs between PV deployment and ecological conservation remain underexplored. This study evaluates the spatial distribution and development potential of PV installations in Yancheng City, China, by integrating deep learning, multi-source data, and ecological constraints. Three deep learning models (UNet, PSPNet, and DeepLabV3+) were applied to identify PV panels on rooftops, water bodies, and bare land. PV maps were overlaid with ecological protection zones to quantify spatial conflicts, and the Analytic Hierarchy Process combined with Entropy-Weighted TOPSIS was employed to assess district-level development potential. UNet achieved the highest Intersection over Union (IoU) for rooftops (97.09%), water bodies (95.78%), and bare land (94.79%). The total PV area reached 71.11 km2, with rooftops accounting for 60.35%, of which 26.56% overlapped ecological protection zones. Areas suitable for future PV development constituted 8.44% of the city, with Dafeng District, Dongtai City, and Tinghu District exhibiting the highest potential. Resource endowment, particularly building coverage (weight = 0.5410), emerged as the dominant factor. This framework provides a replicable approach for integrating ecological constraints into PV potential assessments and supports region-specific planning in urban and peri-urban contexts.