Research on Urban Road Extraction Algorithm of High-Resolution Remote Sensing Images
Roads is often obscured by buildings and trees in high resolution remote sensing images,and easily confused with the background features,which leads to the problem that the extracted roads are easy to be interrupted and distributed in a fragmented manner.In order to solve the above problems,this paper proposed a connectivity-enhanced road extraction network.The network utilized strip convolution to mine continuous features of roads in different directions,and proposed connectivity attention branches based on graphical analysis to mine continuous information between adjacent pixels of roads.In order to realize the road extraction using this algorithm,the original images and road labels were cropped and processed by computer programming,and the connectivity cube was parsed based on the graph structure,and the model was constructed using pytorch and the loss function was used to complete the model training and prediction.In order to test the model precision,this paper carried out road extraction experiments on the Massachusetts dataset and the high resolution image dataset of representative cities in China,and by comparing the extraction results of the model with the comparison models,the recall rate increased by 7.84%,intersection to intersection ratio,and F1 value of the model extraction achieved the highest score,which realized the effective extraction of roads.
NETL
NSTL
万方数据
