面向屋顶光伏规划的建筑物屋顶分割方法

A Building Roof Segmentation Method for Rooftop PV Planning

屋顶分割在屋顶光伏规划中具有重要意义。由于其基于人工提供prompts的方式,降低了屋顶光伏规划的自动化程度。为提高SAM在屋顶分割任务中的自动化水平,本文提出了一种改进Segment Anything Model(SAM)的建筑物屋顶自动分割方法。我们在SAM的基础上引入了一种用于提供prompts的可学习网络。简化了屋顶分割任务,将其转变为端到端的任务,并且显著提高了屋顶分割的准确性。经过实验验证,该方法的分割准确度可达99%,mAP可达82.5%,在建筑物屋顶分割和自动化水平上表现出卓越的性能。

Rooftop segmentation is important in rooftop PV planning.As it is based on manual provision of prompts,it reduces the level of automation in roof PV planning.In order to improve the automation of SAM for roof segmentation tasks,this paper proposes an automated segmentation method for building roofs that improves the Segment Anything Model(SAM).We introduce a learnable network for providing prompts based on SAM.The roof segmentation task is simplified and transformed into an end-to-end task,and the accuracy of roof segmentation is significantly improved.The method is experimentally validated to achieve segmentation accuracy up to 99%and mAP up to 82.5%,demonstrating excellent performance at the level of building roof segmentation and automation.