Study On Seabed Classification Technology Based On Multibeam Bathymetry Sonar

The multibeam bathymetric sonar is main acoustic equipment for remote sensing the seabed characteristics and can provide the full sea depth,wide coverage and high precision seabed topography information.Moreover,the seabed backscatter data received by multibeam bathymetric sonar can be used to distinguish the surficial seabed type.So the multibeam bathymetric sonar can achieve integrated detection of the seabed topography,seabed acoustic imagery,seabed type and other seabed characteristic information,which will improve the efficiency of marine investigation.And how to effectively utilize the acoustic data measured by multibeam bathymetric sonar and reliably distinguish the seabed type are the hot and difficult problems,which have been widespread concerned.This thesis combines the research status and the developing trend of seabed classification technology of multibeam bathymetric sonar,and mainly focuses on the four main aspects of research work,included the seabed backscatter imaging technology of multibeam bathymetric sonar,statistical characteristics of imaging data,feature extraction method of multi-source information and the realization of classification algorithm.The main part of the thesis can be divided into following:1.The seabed backscatter imaging technology of multibeam bathymetric sonar has been studied.This thesis proposes a kind of seabed acoustic imaging method based multibeam interferometric principle.Firstly the basic multibeam interferometric algorithm has been analyzed to estimate the time and direction of arrival(TOA-DOA)of seabed echo,and the phase data quality has been controlled from the interferometric signals amplitude,correlation and spectrum.Basing on these,the echo intensity and spatial location information can be got by using the TOA-DOA data and sound velocity profile in water.The multibeam interferometric imaging algorithm avoids approximation process of snippet method included calculation of the echo intensity sequence position and echo incidence angle,and realizes the accurate integration of the echo intensity and its position,which improves the image quality and has relatively good spatial resolution.Combined with sonar equation,the echo intensity has been corrected to get the backscatter strength data,and this thesis researches on the correct method of angular relationship of backscatter strength to eliminate the angular factor of seabed acoustic image.At last the multibeam interferometric imaging algorithm has been proved by the experiment data.2.The statistical characteristics of the seabed backscatter data from multibeam bathymetric sonar have been studied.Firstly the K distribution model of the seabed backscatter signal amplitude has been analyzed,and the probability distribution of backscatter strength data has been derived,which obeys the K distribution in log domain.Secondly,the applicable conditions and limitations of Gaussian distribution hypothesis for describing the probability distribution of the backscatter strength data have been analyzed.Then this thesis uses the simulation data and the experimental data including a variety of sediment types,two different frequencies to test the above theoretical results,and to verify the correctness of theoretical analysis.At last,the experimental data have been used to analyze the scale and shape parameters of K distribution in log domain,and the test results show that there is a certain relation between the two parameters and incident angle,and there are some differences for the shape parameter and scale parameter at different seabed types.3.The multi-source information feature extraction method and their capabilities of classification have been studied.Firstly,basing on the probability distribution characteristics of data,the texture feature extraction used by gray level co-occurrence matrix,and the Pace features from power spectrum ration,the feature extraction method has been studied.Secondly,the experimental data have been combined to analyze affection of sample window size to classification accuracy.The results showed that with the increase of sample image scale,the classification accuracy is higher,while it increased to a certain extent,the classification accuracy tends to a constant and is no longer affected by the sample window scale.And then,this thesis uses the Fisher discriminant rate to analysis the classification performance of each feature quantity.The two parameters of K distribution in log domain are used as classification features,and have a relatively good classification performance.Then,three groups of feature vectors obtained from the three kinds of feature extraction methods have been tested the classification accuracy used by support vector machine(SVM)classifier and analyzed the classification ability of them.Finally,the angular response curves of backscatter strength data have been used as data source to extract the features,and test their performance using the simulation analysis and experimental data.Compared with the feature vector extracted by the angular response curve and the three groups of feature vectors extracted by the three kinds of extraction methods from seabed acoustic imagery,the classification accuracy is relatively low.And the feature vector obtained by probability distribution characteristics of data is the best,and the accuracy can reach 91.95%for the sample data of five types,included the sand,gravelly sand,sandy gravel,muddy sandy gravel and rock.4.The seabed classification method of multi-source features based on composite kernel SVM has been studied.To maximize the characteristics of variety of feature vector information and improve the classification performance,this thesis uses the composite kernel to combine the different feature vector information instead of single kernel,and uses the SVM to achieve the seabed classification.This thesis discusses the parameters search method based on cross-validation and the evaluation method of classification accuracy,included classification accuracy of the overall sample and Kappa coefficient.Combined with experimental data,the effectiveness of the research methods has been tested.The results show that the seabed classification used by composite kernel SVM can get higher classification accuracy than traditional single kernel form.And the test results indicate that if different feature vector information are directly combined a vector for classification,its classification accuracy does not necessarily higher than the classification accuracy obtained by either one kind of feature vector information,and the composite kernel SVM treatment can effectively solve this problem.