[Purposes]To address the challenges of small-scale and densely distributed rooftop photovoltaic(PV)modules in high-altitude UAV imagery,this study proposes an efficient,automated method for precise PV identification and area estimation to support regional-scale monitoring of distributed PV assets.[Methods]An improved instance segmentation model is developed based on YOLOv8-seg by replacing the original backbone with a lightweight RepViT architecture,utilizing structural re-parameterization to enhance global feature modeling and representation efficiency.Furthermore,a multi-stage feature fusion strategy is employed in the neck network to strengthen the perception of small-scale modules,and a mask-based physical area estimation model is constructed by integrating ground sampling distance(GSD).[Findings]Experiments conducted on UAV datasets from Zengcheng district,Guangzhou,show that the improved model achieves a Box mAP50 of 95.4%and a Mask mAP50 of 94.5%,representing precision increases of 3.4%and 2.9%respectively compared to the baseline.In large-scale practical tests,the model successfully achieved accurate multi-scale photovoltaic area calculation spanning three orders of magnitude,with errors meeting the requirements of engineering applications.[Conclusions]The proposed method effectively reduces parameter count while significantly improving segmentation accuracy and inference speed,providing a high-precision,automated technical solution for the census and fine-grained management of distributed rooftop solar resources.