Extraction Of Rivers And Rivers' Centerline Based On High Resolution Images

The development and perfection of remote sensing technology has made it gradually combined with every walk of life,which puts forward new requirements to research on remote sensing science.High resolution images can provide more details for the extraction of feature to meet the more sophisticated need of application.Aim at the problems of rivers and rivers' centerlines extraction happened in investigating work on rivers,the paper studied how to extract normal rivers and narrow rivers in different background by using domestic high resolution data,and extract and connect rivers'centerline by mathematical morphology and topology.We hope the research can provide some remote sensing technology for water conservancy workers to do the census of rivers.This paper is mainly as follows:1.In combination with domestic high resolution images,analyze the characteristics of rivers in shape,textures,spectrum and neighborhood.Being found through analysis,there are some similarities and differences in normal rivers and narrow rivers.Among the characteristics,spectral index methods are the most commonly used and quickest methods to extract normal rivers.To narrow water,the two characteristics of low value of near-infrared band and obviously change in neighborhood's value which makes narrow rivers presents the typical linear feature characteristics are suitable for extract narrow rivers.2.Research on extracting rivers by using domestic high resolution images,and focus on the extraction of narrow rivers.To normal rivers,use NDWI index methods to extract which is quick in process and applicable relatively widely.To narrow rivers,firstly narrow the scope of the target by the assist of DEM,and then use high pass and near-infrared segmentation to extract narrow rivers in the background of vegetation while use this method after taking reciprocal to near-infrared band in the background of bare rocks.These two kinds of background can represent the background of narrow rivers in eastern China and western China to some extent.Experiment shows that the mentioned method can extract narrow rivers in the specific background,and can maintain the details of narrow rivers.After post-processing,omissive pixels can be supplied partly.3.Extract rivers' centerlines from rivers binary images,and focus on connect the fractures caused by omitted rivers extraction.First extract initial centerlines from rivers binary images.And then extract the break points of initial centerlines and connect river centerlines fracture by dilation-thinning method on break-points only.Finally,remove the wrong connecting lines in topology constraint.Experiment shows that the proposed method can keep the correctness in position and natural semantics.In the proposed method,the connection algorithm conquers the lack of the min pair of breakpoints method that the rate is low while the error rate is high when dealing with large scale images,and also conquers the lack of dilation-thinning method that deformation happens when dealing with long fracture.Removing the wrong connecting line according to topological relation greatly reduces the connection error rate.