摘要
传统建筑是中国宝贵的建筑遗产,承载着优秀的民族建筑文化,是反映城市特色风貌的重要指标。现阶段深度学习识别建筑物的技术相对成熟,但使用街景图片识别建筑遗产并进行地图可视化展示的研究较少,因此,本研究基于Mask-RCNN(mask region-based convolutional neural network)模型,融合SE(squeeze and excitation)注意力机制,提出一种基于SE-Mask-RCNN识别街景图片中建筑遗产的方法。首先,通过路网数据获取百度街景图片,制作数据集。其次,在模型的残差网络(residual network,ResNet)中引入SE注意力机制;并与已有相关方法 U-net(u-shaped network)、全卷积网络(fully convolutional network,FCN)、Mask-RCNN三种模型进行实验对比评价。最后,使用本方法识别研究区域内的街景图片,形成可视化地图,分析建筑遗产在空间上的分布情况。结果表明,本方法可以有效识别城市中的建筑遗产,识别结果较Mask-RCNN、U-Net、FCN模型分别提高了2%、3.1%、4.7%,证明了本方法对城市中建筑遗产的识别具有可靠性和有效性。研究成果可为建筑遗产保护及现状调查提供依据。
The technology for building recognition based on deep learning is relatively mature,but there are limited studies on urban architectural heritage recognition using street view images.This paper introduces a novel architectural heritage recognition method,the SE-Mask-RCNN algorithm,for identifying architectural heritage in the street view image.This method addresses issues related to remote sensing image recognition accuracy and the timeliness of traditional exploration.The results are visually displayed in the space corresponding to the street attractions,effectively determing the location of architectural heritage and the core protection area,thereby providing decision support for architectural heritage preservation.This paper utilizes road network data to crawl street view data based on the characteristics of Chinese architectural heritage.The Mask-RCNN model with the SE mechanism is employed for architectural heritage The recognition results are compared with U-Net,FCN,Mask-RCNN,and other algorithms to analyze the feasibility of the SE-Mask-RCNN algorithm.Additionally,spatial visualization display and nuclear density analysis of the corresponding street attractions are conducted.Experimental results show that the proposed method efficiently and accurately identifies architectural heritage in street view images,with the mAP value of the extraction result being 2%higher than that of Mask-RCNN.The method exhibits favorable characteristics in image recognition results,mainly due to the flatter edge of architectural heritage and lower error rate.Compared with other algorithms,the accuracy and robustness of the method surpass the comparison algorithm.The identified architectural heritage street attractions are mapped onto the road network,reflecting the spatial location of the architectural heritage.From the perspective of spatial distribution,the distribution of architectural heritage in the Xicheng study area shows the following characteristics.①Overall distribution characteristics:analysis of the overall pictur
作者
朱小凡
胡璐锦
王恺
王坚
ZHU Xiaofan;HU Lujin;WANG Kai;WANG Jian(School of Geomatics and Urban Spatial Informatics,Beijing University of Civil Engineering and Architecture,Beijing 102616,China;Gansu Provincial Basic Geographic lnformation Center,Lanzhou 730000,China)
出处
《时空信息学报》
2024年第1期50-56,共7页
JOURNAL OF SPATIO-TEMPORAL INFORMATION
基金
北京建筑大学青年教师科研能力提升计划(X21023)
甘肃省自然资源厅科技创新项目(202252)。
