复杂底质环境下的多波束海底底质分类方法研究

Existed methods for seabed classification based on multibeam measured data are apt to be disturbed by the seabed heterogeneity, which worsen the classification solution. Meanwhile, as far as the class number is concerned, expert judgment accompanied by BIC is involved, which results in a lack of objectivity and accuracy. To account for this, this project plans to study on the unsupervised seabed classification to improve its accuracy and  objectivity. Firstly, we apply the hypothesis test to detect the seabed heterogeneity, and use the Gaussian mixture model(GMM) to extract the back-scatter features of different kinds of seabed subsequently. Secondly, to account for the impact of the incident angle, the back-scatter feature is converted to a multidimensional vector. Meanwhile, GMM is used to learn the obtained vectors. In the end, the number of the class is estimated by analyzing the posterior of the classification result as well as the model complexity, based on statistical theory. As a result of the above research steps, an accurate and objective unsupervised seabed classification method can be acquired. The result of research can improve the seabed characterization technology, which is very important for oceanic scientific research and marine engineering.

目前，使用多波束测量数据进行无监督海底底质分类时，回波特征值易受到非均质海底底质的影响，导致分类结果准确性不足。同时，最优分类数目的选取主要依靠以统计参数为辅助的人工判断，导致其准确性和客观性不足。对此，本项目以基于多波束测量数据的海底底质分类为研究对象，探索客观、准确的无监督海底底质自动分类方法。首先，针对特征提取过程中非均质底质的干扰，使用假设检验方法对其进行探测，并用高斯混合模型提取不同底质的回波特征值。其次，为准确处理波束入射角的显著影响，将回波特征值化为多维向量，并使用高斯混合模型加以表征。最后，在准确提取和表达回波强度特征信息的前提下，基于统计理论，分析不同分类数量所得结果的后验分布函数和模型复杂度，确定最优分类数量的选取方法。计划通过以上研究，形成准确、客观的无监督海底底质自动分类方法。研究成果可促进海底底质探测技术的提高，对海洋科学研究和工程应用有重要意义。

目前，使用多波束测量数据进行无监督海底底质分类时，回波特征值易受到非均质海底底质的影响，导致分类结果准确性不足。对此，本项目以基于多波束测量数据的海底底质分类为研究对象，研究客观、准确的无监督海底底质自动分类方法。为此，项目组通过海试实验收集了南海典型海域（南海甘泉岛、蜈支洲岛）的多波束海底测量数据，以其作为实验数据，并结合仿真数据，开展研究。首先，针对特征提取过程中非均质底质的干扰，使用假设检验方法对其进行探测。其次，为准确处理波束入射角的显著影响，将回波特征值化为多维向量，并使用统计学习模型加以表征。此外，研究发现，在底质分类过程中，针对复杂海底底质环境，同时结合海底地形信息和回波强度信息，进而提取特征向量，可有效提升分类准确率。本研究成果可促进海底底质探测技术的提高，对海洋科学研究和工程应用有重要意义。