RH-CUnet:嵌入边缘和角点的传统村落建筑物提取

RH-CUnet:Extracting Traditional Village Buildings by Embedding Edges and Corners

传统村落建筑物作为珍贵的文化资源,快速、准确地获取传统村落建筑物信息,对保护传统文化有着重要意义。在利用遥感图像识别农村建筑物,尤其是传统村落建筑物时,存在漏检、错检、轮廓失真、角点不规则等问题。针对此问题,文章提出了RH-CUnet模型。该模型以U-Net网络为基础,通过嵌入边缘识别网络和角点检测算法来提高模型对建筑物的边缘、角点的关注度,从而实现传统村落建筑物的精确提取。实验结果表明,在自制的传统村落建筑物数据集和WHU建筑物数据集上,该方法提取的建筑物更加完整,轮廓形状平滑,边角清晰,且在precision、recall、F1-score、IoU、BoundF和VNE 6种评价指标上都得到了很大的提升。RH-CUnet模型有效地提高了传统村落建筑物的提取精度,具有一定的实用价值。

As a precious cultural resource,obtaining information on traditional village buildings quickly and accurately is of great significance for protecting traditional culture.When using remote sensing images for recognizing rural buildings,especially traditional village buildings,it faces problems such as missed detections,false detections,distorted contours,and irregular corners.In response to this issue,this article proposes the RH-CUnet model,which is based on the U-Net network and embeds edge recognition networks and corner detection algorithms to improve the model’s attention to the edges and corners of buildings,thereby achieving accurate extraction of traditional village buildings.The experimental results show that on Traditional Village Building Dataset and WHU Building Dataset,the outline shape of the traditional village buildings extracted by RH-CUnet is smooth and complete,the corners are sharp and clear,and the six evaluation indicators of precision,recall,F1-score IoU,BoundF,and VNE all achieve good results.The RH-CUnet model effectively improves the accuracy of traditional village building land extraction and has certain practical value.