摘要
针对目前已有基于遥感影像道路中心线提取算法易受道路旁树木遮挡、建筑物及其阴影覆盖和道路上车辆等因素影响,造成提取出来的道路中心线存在断裂、不完整现象,提出了一种基于深度学习语义分割的道路掩膜,引用细化算法提取道路中心线矢量数据,对矢量道路中心线进行优化的道路中心线提取方法。首先,通过对深度学习语义分割提取出来的道路掩膜进行形态学膨胀处理,减少道路掩膜出现部分断裂、空洞、不完整现象;然后,利用细化算法,对膨胀处理后的道路掩膜提取道路中心线并进行矢量化;最后,结合出现断裂处的道路中心线间几何、空间等约束关系,进行优化处理。实验结果表明:该方法相对于其他道路中心线提取方法,具有较高的精确度、完整度,在不考虑前期深度学习样本制作、模型训练所使用时间的情况下,提取效率也优于其他方法;生成了标准格式的矢量道路中心线数据,可直接用于实际生产。
In view of the existing road centerline extraction algorithms based on remote sensing images,the extracted road centerline is broken and incomplete due to factors such as the occlusion of roadside trees,the coverage of buildings and their shadows,and the vehicles on the road. In this paper,a road mask based on deep learning semantic segmentation is proposed,which uses a thinning algorithm to extract the vector data of the road centerline,and optimizes the road centerline extraction method of the vector road centerline. Firstly,morphological expansion is performed on the road mask extracted by deep learning semantic segmentation to reduce the occurrence of partial fractures,holes,and incompleteness in the road mask. Secondly,the thinning algorithm is used to extract the road center line from the expanded road mask and carry out vectorization. Finally,combine with the geometric and spatial constraint relations between the center lines of the road where the fracture occurs,and perform optimization processing. The experimental results show that compared with other road centerline extraction methods,the method in this paper has higher accuracy and completeness of the extracted road centerline. The extraction method does not consider the time used for the preparation of early deep learning samples and model training. The efficiency is also better than that of other methods,and vector road centerline data in a standard format is generated,which can be directly used in actual production.
作者
张荣军
谭海
马天浩
于永帅
黄小贤
ZHANG Rongjun;TAN Hai;MA Tianhao;YU Yongshuai;HUANG Xiaoxian(School of Geomatics,Liaoning Technical University,Fuxin,Liaoning 123000,China;Land Satellite Remote Sensing Application Center,MNR,Beijing 100048,China;Airborne Survey and Remote Sensing Center of Nuclear Industry,Shijiazhuang 050000,China)
出处
《遥感信息》
CSCD
北大核心
2022年第1期94-100,共7页
Remote Sensing Information
