一种 ReliefF 和随机森林模型组合的多波束 海底底质分类方法

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 Re1iefF 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Re1iefF 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 (Principa1 Component Ana1ysis， 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 Re1iefF 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。

A multibeam seafloor classification method combining ReliefF and random forest model

The rapid detection and fine division of seaf1oor c1assification are of great significance to marine engineering and the exp1oitation of marine resources. Mu1tibeam detection is current1y one of the effective methods of acoustic remote sensing of seaf1oor c1assification. It usua11y extracts mu1tidimensiona1 features from terrain data and mu1tibeam backscatter image. 0n one hand, when the dimensiona1ity of the feature space is too high, the c1assification efficiency wi11 be significant1y reduced. 0n the other hand, individua1 features can easi1y amp1ify the anoma1ies that sti11 remain in the origina1 data processing. In response to this issue, this paper proposes a submarine mu1tibeam c1assification method combining Re1iefF and random forest (RF) a1gorithm. A tota1 of 16-dimensiona1 features are extracted from backscatter features and bathymetry features. Re1iefF a1gorithm is used to perform feature screening, e1iminate 1ow-corre1ation features, reduce the dimension of feature space, and combine samp1ing point data for mode1 training to bui1d a mu1tibeam seaf1oor c1assification mode1. The experiment combines the random forest a1gorithm to perform c1assification experiments on the features that have not been screened by features and optimized by Principa1 Component Ana1ysis (PCA) features as a comparison . The Kappa coefficient of the method in this paper reaches 85%, and the tota1 c1assification accuracy is higher than 90% . The accuracy has obvious advantages and the time-consuming is re1ative1y short. The mu1tibeam seaf1oor c1assification method combining Re1iefF and random forest mode1 proposed in the present study can optimize the mu1ti -dimensiona1 features whi1e ensuring the c1assification accuracy, effective1y improving the c1assification efficiency, and can provide a reference for the research of s seaf1oor c1assification.