一种ReliefF和随机森林模型组合的多波束海底底质分类方法  

A multibeam seafloor classification method combining ReliefF and random forest model

作　　者：孙健[1,3] 樊妙 崔晓东[1,3] 艾波[1] 马丹 阳凡林[1,3] SUN Jian;FAN Miao;CUI Xiaodong;AI Bo;MA Dan;YANG Fanlin(College of Geodesy and Geomatics,Shandong University of Science and Technology,Qingdao 266590,China;National Marine Data and Information Service,Tianjin 300171,China;Key Laboratory of Ocean Geomatics,Ministry of Natural Resources of China,Qingdao 266590,China.)

机构地区：[1]山东科技大学测绘与空间信息学院,山东青岛266590 [2]国家海洋信息中心,天津300171 [3]自然资源部海洋测绘重点实验室,山东青岛266590

出　　处：《海洋通报》2022年第2期131-139,共9页Marine Science Bulletin

基　　金：国家自然科学基金(41930535);国家重点研发计划(2018YFC1405900);山东省研究生教育创新计划建设项目(SDYJG19083);山东科技大学科研创新团队支持计划(2019TDJH103)。

摘　　要：海底底质的快速探测和精细划分对海洋工程建设、海洋资源开发等具有重要意义。多波束探测是目前声学底质遥测的有效手段之一,通常提取多波束反向散射强度图像和地形数据中的多维特征结合分类器进行底质分类。一方面,若特征空间维数过高,分类效率会显著降低;另一方面,个别特征容易放大原始数据处理过程中仍存留的异常现象。针对这一问题,本文提出了一种结合ReliefF算法和随机森林(Random Forest,RF)算法的多波束底质分类方法。提取反向散射强度和地形共16维特征,利用ReliefF算法进行特征筛选,排除低相关性特征,降低特征空间维数,结合采样点数据进行模型训练以构建多波束底质分类模型。试验结合随机森林算法对未经特征筛选、经主成分分析(Principal Component Analysis,PCA)特征优化后的特征进行分类实验作为对比。本文方法 Kappa系数达到85%,分类总精度高于90%,精度具有明显优势,耗时也比较短。可见,本文提出的结合ReliefF和随机森林模型的多波束底质分类方法可以在保证分类精度的同时对多维特征进行优化,有效地提高了分类效率,可对海底底质分类研究提供参考。The rapid detection and fine division of seafloor classification are of great significance to marine engineering and the exploitation of marine resources. Multibeam detection is currently one of the effective methods of acoustic remote sensing of seafloor classification. It usually extracts multidimensional features from terrain data and multibeam backscatter image. On one hand, when the dimensionality of the feature space is too high, the classification efficiency will be significantly reduced.On the other hand, individual features can easily amplify the anomalies that still remain in the original data processing. In response to this issue, this paper proposes a submarine multibeam classification method combining ReliefF and random forest(RF) algorithm. A total of 16-dimensional features are extracted from backscatter features and bathymetry features. ReliefF algorithm is used to perform feature screening, eliminate low-correlation features, reduce the dimension of feature space, and combine sampling point data for model training to build a multibeam seafloor classification model. The experiment combines the random forest algorithm to perform classification experiments on the features that have not been screened by features and optimized by Principal Component Analysis(PCA) features as a comparison. The Kappa coefficient of the method in this paper reaches 85%, and the total classification accuracy is higher than 90%. The accuracy has obvious advantages and the time-consuming is relatively short. The multibeam seafloor classification method combining ReliefF and random forest model proposed in the present study can optimize the multi-dimensional features while ensuring the classification accuracy,effectively improving the classification efficiency, and can provide a reference for the research of s seafloor classification.

关 键 词：RELIEFF 随机森林 底质分类 反向散射强度图像 地形特征 

分 类 号：P717[天文地球—海洋科学] P229.1