Multi-scale Convolutional Neural Network For Road Extraction In Remote Sensing Imagery

High resolution remote sensing image is a comprehensive reflection of the earth’s surface features.It analyzes and processes the information in the image through image analysis and information extraction,so as to serve the geographic information system,smart city construction,intelligent transportation system and other fields.With the development of economy and the increase of population,urban traffic has become convenient but crowded.Therefore,the extraction and analysis of the road is a meaningful work.In the face of massive and complex remote sensing data,how to extract the road quickly and effectively is an urgent problem to be solved.Deep learning has obtained a great success in image classification,since it can directly learn from labeled training samples and extract different level image features to encode the input image.In this paper,a deep and systematic analysis on how to automatically extract roads from remote sensing images is carried out.The previous research results are analyzed,and we design a new road extraction method.The main work are summarized as follows:1)In this paper,the research status of road extraction in remote sensing image is summarized,and the advantages and disadvantages of traditional road extraction methods and deep learning methods are analyzed.This paper analyzes the difficulties in road extraction from high-resolution remote sensing images,and expounds the data characteristics of roads in remote sensing images.We propose a multi-scale convolutional neural network(MSCNN)for extracting road from high-resolution remote sensing image,in which road detection can be seen as a regional classification.This core trainable detection engine consists of an encoder-decoder network and a fusion model.Firstly,image was encoded as a feature representation with several stacked convolutional layers.Then,the pre-trained decoder network outputs a series of classification maps guided by the different scale road training data.Finally,we investigate the fusion model utilizing different scale classification maps and obtain a final road decision map.2)The results of road extraction using convolutional neural network sometimes lead to breakage and noise patches.In response to this phenomenon,this paper uses the morphology algorithm to further optimize the results of road extraction.Firstly,the morphological opening and closing operation are used to connect the road breaks;then we based on geometric characteristics of the road,the noise patches in the extracted results were removed to achieve a simple and effective post-processing effect.On the two common high resolution remote sensing image datasets,the proposed road extraction method is compared with several representative methods in the field of deep learning.Experiments shows that the proposed road extraction method for remote sensing imagery can extract the road effectively,and there is a certain improvement compared with other methods in the extraction.