Research On Road Extraction Method Based On FCN And Seed In Remote Sensing Images

High resolution remote sensing image of road recognition is one of the hot spot of current research.Along with the high-speed development of economy and the rapidlyincrease of the urban population.Traffic jams gradually become one of the modern urban problems.By the extraction of real time remote sensing image can get the information of the congested sections roads effectively,what make a difference of roads' extraction.Firstly,using FCN to extract road,the results are abstracted into several road blocks.In order to extract accurate roads,these road blocks are entered to the algorithm that adapt the idea of seed filling.The method is applied to the Massachusett satellite images,the results show that this method has a good effect on the integrity and accuracy of the path recognition under complex environment of cities,towns,interchange,etc.The main content and innovation of this paper is:This paper expounds the principle and method of deep learning in image semantic segmentation.By the concept of deconvolution and traditional network CNN,thenadjust the network so that to applied to the roads' extraction.First of all,this article deals the remote sensing images with the idea of FCN semantic segmentation,to extract the road and the approximate location of the house.It serves as the precondition of the algorithm.To the multiple unconnected roads in an image,when using FCN to process image for roads' extraction,this method has a lower accuracy,at the same time,extraction results of continuous road are not continuous.However,the result can be distributed on the discontinuous roads,besides,this method also has very strong generalization ability.Aiming at the problem of poor generalization ability and artificial seed input in the semi-automatic road extraction algorithm.This paper is based on the advantages of the above method,use the extracted roads and buildings as premise of the algorithm for further processing.Secondly,this paper proposesa methodusing the idea of seed filling to extract road.Aiming at the problem of the part of road segmentsis not extracted in the segmentation resultand the width of road is not accurate,the solutiont of his article is the thought of seed filling and gradually driven search.This paper has proposed an effective method which can replace the search and judgment process in seed filling algorithm.The idea of seed filling in connected region is applied to the extraction process of roads.By the way of judgment,this paper confirm the connected position instead of the known connected region in the seed filling algorithm.By means of search several target image blocks at a certain angle,to instead the search of four directions of the four-connected.On the whole,the boundary of the road is regarded as the boundary in the seed filling algorithm.A section of the road is regarded as a connected area.According to the series of judgments,my method can extract the exact road width of the roads.In order to improve the efficiency of the search and narrow the scope of the search,according to the characteristics of the direction of the roads,this paper has proposed a directional search method that obtain satisfactory results in the process of road extraction with seed filling.