Partition Correction Method Of Road Vector Data Supported By Deep Learning

As an important part of geographic information database,road vector data plays an important basic role in social and economic development,and it is an important part of special data such as smart cities and disaster relief.Therefore,scholars have carried out a lot of research in this field.However,due to the significant spatial heterogeneity of urban roads and rural roads,the existing methods are vulnerable to the limitation of regional scene changes,and there is a problem of low universality.Aiming at this problem,this paper proposes a road vector data partition correction method supported by deep learning,this method completes road vector data correction by analyzing the characteristics of urban roads and rural roads,based on deep learning methods and geographical partitioning ideas,including:(1)Feature set extraction.In order to establish the data source of the vector correction method in this paper,firstly,U~2-Net is used to extract urban and rural roads to complete the task of cross-regional road extraction,avoiding the problem of poor universality of traditional road extraction;secondly,using the line segment sequence method to achieve irregular geometric representation of ground objects to obtain the continuity information of road edges;finally,combining the results of deep learning road extraction and line segment sequence extraction,according to the characteristics of urban roads and rural roads,establish urban road feature sets and rural road features set as the data basis for vector correction in this paper.(2)A correction method of urban road vector is proposed.In view of the characteristics of poor quality of urban road deep learning extraction results,poor road image structure information,and regularization of the geometric shape of vector data,the vector line is decomposed at the sudden change of urban road curvature,which simplifies the shape of the vector line and obtains the sub-vector line;according to the characteristics of the road feature set,a random average sampling method is proposed to fit and correct the sub-vector lines;in addition,in order to make the position of the vector lines more in line with the road centerline,the sub-vector lines are centralized;finally,the fitting vector lines are connected,complete the urban road vector data correction.(3)The vector correction method of rural road is proposed.In view of the characteristics of high quality extraction results of rural roads,strong road image structure information,and irregular geometry of vector data,based on the narrowness and length of rural roads,the geometric constraint method is used to obtain the optimal road edge line in the rural road feature set.On this basis,the road prior knowledge is obtained from the vector,the road centerline inference model is established,the road centerline is extracted,and the vector correction is done indirectly;finally,analyze the reasons for the rupture of rural road vector lines,and complete the correction of rural road vector data by dynamically selecting connecting vector lines.In order to verify the universality of the method,this paper selects two urban and rural remote sensing images as experimental data,and compares them with the existing vector data correction methods.The experimental results show that the method in this paper has the advantages of strong universality and high accuracy on the basis of ensuring that the vector data can be stored in the database.