引用本文:	兰志刚,靳卫卫,朱明亮,于新生,国建凤,周振涛,李凯宝.基于高斯混合模型的海冰图像非监督聚类分割研究[J].海洋科学,2011,35(11):97-100.

【打印本页】【HTML】【下载PDF全文】【查看/发表评论】【下载PDF阅读器】【关闭】

过刊浏览高级检索

本文已被：浏览 2067次下载 2951次	码上扫一扫！
分享到：微信更多字体:加大+\|默认\|缩小-
基于高斯混合模型的海冰图像非监督聚类分割研究
兰志刚¹, 靳卫卫², 朱明亮², 于新生², 国建凤³, 周振涛⁴, 李凯宝⁴
1.中海油能源发展股份有限公司北京分公司;2.中国海洋大学海洋地球科学学院;3.中国海洋石油总公司;4.中海石油有限公司天津分公司

摘要:

为了利用海冰图像识别技术获取海冰冰况信息, 探索了利用高斯混合模型进行海冰图像分割的技术途径, 描述了具体算法, 并利用高斯混合模型的最大期望值(EM)算法以及最小描述长度(MDL)准则对渤海海冰图像进行目标提取。研究结果表明, 该方法可以很好地实现海冰信息的有效提取和海冰图像的有效分割, 从而证明了建立在图像分割技术之上的海冰图像识别技术是处理海冰图像进而获得冰型、冰量等冰况信息的有效技术手段。

关键词: 海冰高斯混和模型图像分割非监督聚类

DOI：

分类号:

基金项目:中国海洋石油总公司科技发展项目(C/KJFJDSY 003-2008)

Sea ice image segmentation with unsupervised clustering based on the Gaussian mixture model

Abstract:

In order to obtain sea ice data from in situ video images, sea ice images were processed with image segmentation technology based on the Gaussian mixture model (GMM). Image segmentation of the Bohai sea ice with unsupervised clustering was realized by the expectation-maximization (EM) algorithm of GMM and minimum description length (MDL) criterion on the sea ice images for object extraction. The calculation procedures of sea ice image segmentation was described. The results indicate that GMM is effective in sea ice image segmentation and sea ice data extraction. It is concluded that sea ice image recognition, based on image segmentation, is an effective technology to process sea ice image for extraction of data on sea ice type and abundance.

Key words: sea ice, Gaussian mixture model, image segmentation, unsupervised clustering