一种网箱养殖区自动提取方法

An Automatic Extraction Method for Cage-fenced Culture Areas

网箱养殖区的动态监测与分析是水库环境监测的一项重要指标。鉴于网箱养殖区在影像中存在较弱的光谱和形状规律性,自动提取难度较大,提出了一种综合基于像元和面向对象优势的自动提取方法。首先,通过优化归一化差分水体指数(optimized normalized difference water index,ONDWI)将水体(包含网箱养殖区)与陆地分离;然后,利用基于比值指数的纹理特征分析法,将网箱养殖区(包含小岛及其他干扰项)与自然水体分离;最后,基于简化的区域分割算法,同时顾及网箱养殖区的形状特征将其与小岛及其他干扰项分离,最终实现网箱养殖区的精准提取。实验结果表明,该方法能够较为完整地提取出全部网箱养殖区,提取精度为88.1%,高于最小距离法(73.8%)和最大似然法(42.9%),且较好保留了网箱养殖区规则的形态特征。

The dynamic monitoring and analysis of cage-fenced culture areas is an important index of reservoir environment monitoring.Since it is a challenge to automatically extract cage-fenced culture areas because of weak spectra and shape regularity in images,an automatic extraction method combining pixel-based and object-oriented techniques is proposed.Firstly,water body is separated from land by using the optimized normalized difference water index(ONDWI).Then,cage-fenced culture areas including island and other interference are separated from natural water by using the texture feature analysis method based on ratio index.Finally,cage-fenced culture areas are separated from island and other interference by using simplified domain division method and considering the regular shape of cage-fenced culture area.The experimental results show that the proposed method can extract the cage-fenced culture areas completely,with an extraction accuracy of 88.1%,higher than the minimum distance method(73.8%)and the maximum likelihood method(42.9%).Furthermore,it retains the regular morphological characteristics of cage-fenced culture areas.