多波束测深声呐海底底质分类技术研究

发布时间：2018-01-03 01:19

本文关键词：多波束测深声呐海底底质分类技术研究　出处：《哈尔滨工程大学》2014年博士论文　论文类型：学位论文

【摘要】：多波束测深声呐是海底特征声学遥感的主流设备之一,不仅能提供全海深、宽覆盖、高精度的海底地形地貌信息,而且利用其接收的海底反向散射数据还可用于判别海底表层底质信息,从而能实现海底地形、地貌与底质类型多种海底特征信息的一体化探测,这有利于提高海洋调查的工作效率。而如何有效地利用多波束测深声呐测量的声学数据并可靠地判断海底底质类型长期以来一直是人们广泛关注地热点与难点问题。本文结合国内外多波束测深声呐海底底质分类技术的研究现状与发展趋势,重点围绕具有完全自主知识产权的国产多波束测深声呐海底反向散射成像,成像数据的统计特性,多源信息特征提取方法以及分类算法的实现等4个主要方面开展研究工作,并以理论分析、计算机仿真以及湖上与海上试验数据处理等理论和实验方法为研究手段,主要研究内容如下:1、研究基于多波束测深声呐的海底反向散射成像技术。在总结分析现有3种基于多波束测深声呐的海底成像方法原理及其各自优缺点的基础上,提出了一种基于多波束相干原理的海底声学成像方法。首先理论研究多波束相干算法估计海底回波到达时间(Time of Arrival,TOA)与到达角度(Direction of Arrival,DOA)信息的基本方法,并在估计相位差序列时为抑制噪声影响,从相干信号的幅度、相关性以及频谱3个角度对相位差数据进行质量控制。在此基础上,利用得到的TOA-DOA数据并考虑水中声速对声波传播路径的影响对海底检测点的回波强度以及空间位置进行估计。由于该成像方法在对海底检测点的回波强度及其位置进行测量时都利用共同的TOA-DOA信息,因此避免了 snippet法中对各波束内回波强度序列的位置及其入射角度计算时的近似处理,实现回波强度及其空间位置数据的准确融合,提高了成像质量,并且该方法具有良好的空间分辨率。然后,结合声呐方程对回波强度数据进行修正,获取与海底底质特征联系更为紧密的反向散射强度数据,并分析研究反向散射强度的角度关系修正模型,以剔除其对海底声图像显示的不利影响。最后,通过试验数据对基于多波束相干原理的海底成像方法性能进行检验,验证了该方法的有效性。2、对多波束海底反向散射数据的统计特性进行研究。在理论分析海底反向散射信号幅度服从K分布模型的基础上,推导得到海底反向散射强度数据的概率分布服从对数域K分布,并估计模型参数的表达式。通过对上述概率分布模型的进一步近似推导,分析高斯分布假设描述反向散射强度概率分布的适用条件与局限性。然后利用仿真数据以及多种底质类型、两个不同频率的多波束测深声呐试验数据对上述理论结果进行检验,验证了理论分析的正确性和实用性。最后对不同底质、不同角度下反向散射强度数据的对数域K分布模型两参数——尺度参数与形状参数的一般变化规律进行分析,试验结果表明两参数与入射角度之间存在一定联系,且不同底质下的形状参数与尺度参数存在一定差别。3、对多源信息特征提取方法及其分类性能进行研究。首先以多波束海底声图像为数据源研究基于数据概率分布特性的特征提取方法,基于灰度共生矩阵的纹理特征提取方法以及基于功率谱比的Pace特征提取方法;其次,结合试验数据分析图像样本窗大小对分类结果的影响,结果表明随着样本图像尺度的增加,分类正确率越高,而当增大到一定程度时分类正确率趋于恒定,不再受样本窗尺度的影响;并且通过利用Fisher判别比对上述特征提取方法得到各特征量的分类性能进行比对分析;其中,将对数域K分布两参数用作分类特征量,且分类性能较好;然后,将各特征提取方法得到的特征量构成3组特征向量,并利用支持向量机(Support Vector Machine,SVM)分类器对各特征向量整体的分类能力进行综合分析,最后,以反向散射强度数据的角度响应曲线为数据源提取分类特征,并对其性能进行仿真分析以及试验数据的检验。从整体上看利用角度响应曲线提取的特征向量与上述3种基于声图像的特征提取方法相比,分类正确率相对较低,其中基于数据概率分布特性得到的特征向量分类性能最好,对砂、砾质砂、砂质砾、泥质砂质砾以及岩石5种类型样本数据的总体分类正确率可达到91.95%。4、研究多源特征合成核SVM的多波束底质分类方法。为使多种特征信息联合使用后充分发挥各自特点,提高分类性能,本文在理论分析传统单核SVM分类器分类原理的基础上提出利用合成核SVM进行多波束海底底质分类的方法,即将不同特征信息数据以加权加法形式构成合成核以代替传统的单核形式,并用SVM分类器进行海底底质分类。讨论了分类算法中最优参数的交叉验证搜索方法以及总体样本正确率、Kappa系数等分类正确率评价方法。并在此基础上,结合试验数据对本文研究方法的有效性进行检验。试验结果表明,基于合成核SVM的海底底质分类可得到比传统单核SVM更高的分类正确率,验证了利用合成核SVM在该分类问题中的有效性;并且试验结果表明,不同特征信息联合使用后,如果直接合成一个向量进行分类,其分类正确率并不一定能比单独一种特征信息获得的分类正确率高,而经合成核SVM处理后可有效解决此问题。
[Abstract]:Multibeam sonar is one of the main characteristics of the acoustic remote sensing equipment, can not only provide the deep sea, wide coverage, seabed topography information with high accuracy, and the use of the seafloor backscatter data received can be used for distinguishing seabed material information, which can realize the integration of the seabed topography, landform and sediment detection the characteristics of the various types of information, which is conducive to improve work efficiency. The marine survey and how to effectively use of acoustic data in multi beam bathymetry sonar measurement and reliably determine the seabed types has long been widespread attention to hot and difficult problems. Combining with the present situation and development trend of domestic multi beam bathymetry sonar seabed classification technology, it has completely independent intellectual property rights of the domestic multi beam bathymetry sonar seafloor backscatter imaging, imaging data system Project characteristics, to carry out research work in 4 aspects of multi-source information extraction method and classification algorithm, and based on theoretical analysis, computer simulation and lake and sea test data processing theory and experimental method as the research method, the main research contents are as follows: 1. Research on the seafloor backscatter imaging sonar multibeam sounding based on. Based on summarizing and analyzing the existing 3 kinds of principle of multi beam bathymetry sonar seafloor imaging method based on its advantages and disadvantages, proposes a method of multi beam coherent underwater acoustic imaging. Based on the principle of the first theoretical study of multi beam coherent algorithm to estimate seabed echo arrival time (Time of, Arrival, TOA) and angle of arrival (Direction of Arrival, DOA) the basic methods of information, and in the estimation of phase difference sequence for noise suppression, the coherent signal amplitude, correlation and spectrum 3 A view of the phase difference data quality control. On this basis, using TOA-DOA data and considering the influence of water velocity on the wave propagation path of the echo intensity of submarine detection point and location estimation. Due to the imaging method for measuring the seabed detection point back wave intensity and position are the common TOA-DOA information, so as to avoid the approximate position and angle of incidence on the echo intensity of each beam in sequence by snippet method, to achieve accurate fusion of echo intensity and spatial data, the image quality is improved, and the method has good spatial resolution. Then, combined with the sonar equation on the echo data are corrected. The acquisition and seabed features more backscatter data closely, and analyze the relationship between backscattering intensity correction angle In order to eliminate the model of acoustic image shows adverse effects. Finally, through the test on the performance of underwater multi beam coherent imaging method based on the principle of test data, verified the effectiveness of the method was.2, the statistical characteristics of multibeam seafloor backscatter data research. In the theoretical analysis of seafloor backscattering amplitude obey the basic K distribution model, derived the probability distribution of the seafloor backscatter strength data obey logarithmic K distribution, and the expression to estimate model parameters. Through further approximation of the probability distribution model is derived, analysis of the Gauss distribution describes the backscatter intensity probability distribution of the applicable conditions and limitations. Then using the simulation data as well as a variety of sediment types, to test the above theoretical results are two different frequency multibeam sonar test data, verify the theoretical analysis. The correctness and practicability. The different substrate, different angles backscatter data of the log domain K distribution model of two - parameter scale parameter and shape parameter changes in general analysis, test results show that there is a certain relation between the two parameters and incident angle, there are some differences and.3 under different substrate shape parameter and scale parameter, the multi-source feature extraction method and its classification performance is studied. Firstly, through multi beam acoustic image feature extraction method based on probability distribution characteristics of data on the data source, the texture feature of gray level co-occurrence matrix extraction method and extraction method of power spectrum based on Pace characteristic ratio; secondly, combined with the analysis of the test data effect of window size on the sample image classification results, the results show that with the increase of sample image scale, the correct classification rate is higher, and when increased to To a certain extent when the correct classification rate tends to be constant, is no longer affected by sample window scale; and the classification performance by using Fisher discriminant feature extraction method comparing the various characteristic parameters were compared and analyzed; the logarithmic domain K distribution of two parameters for the classification features, and better classification performance; then, 3 feature vectors constitute characteristic quantity method will extract the feature, and the use of support vector machine (Support Vector Machine, SVM) classifier to each feature vector of the overall classification ability of comprehensive analysis, finally, the backscatter data point response curve extracted features as the data source, and tested the simulation analysis and test data of the look at the response performance. The feature vector extraction curve and above-mentioned 3 kinds of extraction methods compared based on features of acoustic images using the angle from the whole, the correct classification rate is relatively low The classification, performance data distribution characteristics based on the best of sand, gravel, sand, sandy gravel, sandy gravel and mud in the overall classification of 5 types of rock sample data accuracy can reach 91.95%.4, the multi beam seafloor classification method of multi-source feature synthesis of nuclear SVM. Give full play to their respective characteristics the combined use of multiple features, improve the classification performance, this paper based classification principle of traditional single kernel SVM classifier is proposed on the use of synthetic nuclear SVM multi beam seafloor classification method in the theoretical analysis, the different forms of single core feature information data in the form of weighted addition to replace the traditional nuclear synthesis, and seabed classification using SVM classifier. The optimal parameters of cross validation classification algorithm in the search method and overall sample accuracy, Kappa coefficient classification accuracy evaluation method. And on this basis, verify with the test data of the research methods of this paper. The experimental results show that the synthesis of nuclear SVM seabed classification can get higher classification than traditional single core SVM accuracy based on validation of the use of nuclear SVM synthesis in the classification of the validity of the test results; and show that the combined use of different feature information, if the direct synthesis of a vector classification, the classification accuracy is not necessarily than a single kind of feature information to obtain the correct rate, and the synthesis of nuclear SVM treatment can effectively solve this problem.

【学位授予单位】：哈尔滨工程大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：P714

【相似文献】