基于决策树与密度聚类的高分辨率影像海岸线提取方法

针对高分辨率影像进行海岸线提取时常常出现的河道区域岸线向陆地方向伸向较远的问题，本文将海岸线视为海水（包括潮滩）与陆地（包括河流）两类地物的分界线进行提取。首先采用C4.5决策树分类方法进行海岸带地物分类规则挖掘，实现基于规则的地物分类；再对海水与陆地分类结果进行基于密度的聚类方法进行后处理，实现噪声去除，其基本原理为:设置一邻域半径，通过统计半径内异类样本点的数量来确定当前点是否为噪声点，若异类像素点的个数超过某一预设的阈值，则对当前噪声点进行修正。为验证本文提出方法的有效性，获取了2013年10月20日天津附近海岸带区域的资源三号卫星影像数据进行验证，结果表明，本文提出的岸线提取方法能够消除河道区域岸线提取的影响，除个别地物比较复杂的区域之外，其平均提取精度优于2个像元，满足海域遥感技术规程中线状信息误差标准的要求。

Coastline extraction from high-resolution image based on decision tree and density clustering algorithms

When extracting coastline from high resolution images, the coastline often extends landward far away due to the influence of the rivers or ditches.In view of this problem, this paper regards coastline as the boundary of seawater (including tidal flat) and land (including rivers and ditches), and then extracts it from images.Firstly, C4.5 decision tree classification method is used to discovery the classification rules for distinguishing the coastal land uses or marine uses; secondly, the discoveried rules are used to classify the coastal land uses or marine uses; after that, postprocessing is conducted based on the classification results by using density clustering method.For a pixel, the principle of postprocessing is to correct the pixel's class according to statistic the number of heterogeneous pixels within a pre-setting radius arround the pixel.If the number of heterogeneous pixels arround exceeds a preset threshold, the current pixel's class will be corrected to the class of the heterogeneous pixels.To validate the proposed method, Resources three satellite image on October 20, 2013 near Tianjin is acquired.The experiment result shows that the proposed method can eliminate the influence of rivers and ditches.The average accuracy is better than 2 pixels, which meets the requirements of the standard of linear information error in the technical regulation of marine remote sensing, except for the region, in which some individual objects are more complex.