Research On Road Extraction Of Remote Sensing Image Based On Convolutional Neural Network

With the rapid development of computer technology and remote sensing technology,high-resolution remote sensing images have become an important information source with high precision,high spectrum and wide coverage.However,how to extract the object quickly,efficiently and intelligently from the massive images is a matter of interest and importance to the remote sensing community.It is of great significance to identify and extract the roads for GIS data updating,map drawing and path analysis.It is difficult to extract roads from high-resolution remote sensing images with noise and complex natural scenes(such as houses,trees)and other factors.Domestic and foreign scholars have made a lot of research on road extraction methods,but there is still no effective approach to extract the road.So,it is very important to study how to extract roads from high-resolution remote sensing images effectively,quickly and intelligently.In this paper,there is a deep and systematical research on the automatic extraction of road from high-resolution remote sensing images.An approach to automatically extract the road is proposed.The details of this paper are as follows:Firstly,the characteristics of remote sensing technology and high resolution remote sensing images are summarized,which lays the theoretical foundation for the research of road extraction method.Then the paper analyzes the advantages and disadvantages of convolution neural network,and proposes a two-classification idea of high-resolution remote sensing image by using convolution neural network to extract road information.Combining with the characteristics of convolution neural network model,we try to improve it from the aspects of training algorithm and cost function.Due to the influence of the natural scene factors such as house and tree shadows,there is still a lot of non-road object noise in the road result,so the result of road feature extraction is further optimized by shape feature analysis and mathematical morphology algorithm.In this paper,some high-resolution remote sensing images are selected for experiments.The experimental results show that the new method can extract the road accurately and completely,and it is suitable for the extraction of roads in the high-resolution remote sensing images with rich details.