A Fast Rooftop Extraction Deep Learning Method based on PP-LiteSeg for UAV Imaginary

Rapid and accurate extraction of rooftops from unmanned aerial vehicle images plays a vital role in many applications, such as automatic photovoltaic resources, urban planning, disaster Risk management, and infrastructure monitoring. A new fast rooftop extraction method based on PP-LiteSeg for UAV images was proposed. Primarily, the input UAV image is preprocessed to enhance the contrast and normalized the images into two classification one is roof top extraction and second background of the image. Then, a deep learning super pixel segmentation algorithm called PP-LiteSeg is employed to generate super pixel regions, which group pixels with similar characteristics and facilitate subsequent processing steps. Next, PaddlePaddle-based classification model EISeg(Efficient Interactive Segmentation) is trained using a large dataset of labeled UAV roof top buildings images which include building and rooftop of building, where each super pixel region is assigned, a binary label indicating whether it belongs to a rooftop or not. Finally, experimental evaluations on a wide-ranging dataset of UAV images prove the effectiveness and efficiency of the proposed method. The results show that the proposed method achieves high accuracy in rooftop extraction, while significantly reducing computational time compared to existing methods. In conclusion, the proposed fast rooftop extraction method based on PP-LiteSeg for UAV images offers a valuable solution for rapid and reliable rooftop of building identification in Shaoguan area. Its combination of super pixel segmentation and PaddleSeg-based image segmentation provides accurate results while minimizing computational complexity. The method's usefulness, effectiveness, and robustness make it a promising implement for applications needful rooftop extraction from UAV imagery.