Fusing And Classifying Multi-beam Sonar And Side-scan Sonar Data

In 21st century, rapidly developed technology make Hydrography go up a higher stage. Many new observing methods have been used in oceanography. Multi-beam echo sonar system (MBES) firstly showed in 70s last century. It developed based on echo sounder. In a ping, hundreds of water depth data can be obtained in acrosstrack by MBES. After a track completed, it can get one swath depth data. The bathymetry swath width can be long as four times as bathymetry, so it can be applied in finding object, rapid mapping and high-resolution depicting seafloor terrain. At the same time, it can also obtain backscattering intensity like side scan sonar system (SSS). SSS is a towed instrument in long-range, high-resolution seabed mapping. It is close to seafloor. It records acoustic backscatter amplitudes vs. time to produce digital SSS image. Some advanced SSS can measure arrival angle of the acoustical returns vs. time using two separate transducers, and convert the angle into bathymetry. The acoustic intensities on each side of the tow-vehicle will be transformed into grey pixel. One ping can obtain thousands of grey pixels. Compared SSS to MBES, the former has high-resolution backscatter image and bathymetry, but poor horizontal position accuracy; the latter has accurate Bathymetry and horizontal position, but lower resolution. So MBES and SSS can best complement each other. In order to fully utilize all information, it is needed to fuse their image and data. And image co-registration is important and complicated work before fusing. This dissertation mainly discusses data processing and fusing multi-beam sonar and side-scan sonar data. It includes several main processes as follows:Firstly, the dissertation introduces the theory and method of instrument operation.Secondly, the dissertation introduces the method of measuring and filtering tide. It emphasizes GPS Tides. GPS Tides is a new method, which adopts GPS RTK technology during the measuring of tide. If the height of the phase center of GPS antenna on the vessel is known, the tide can be calculated. It is an important technical innovation in tidal observation and GPS technology application. On water, especially on ocean, wind and wave will cause the change of attitude of vessel. In this situation, errors will be added to GPS tidal observation data, so the effect of attitude should be eliminated. According to character of wave energy, the error caused by the attitude is discussed respectively. After determining the carrier attitude, the effect of carrier attitude can be eliminated according the attitude data. Because frequency of wave is far higher than the frequency of tide, Wavelet Analysis is used to separate tide from the instantaneous sea surface. After eliminating these errors, the centimeter level tide can be gotten.Thirdly, the dissertation introduces the history and some algorithms in bathymetry filtering. Advances in technology and more indispensable requirements for hydrographic survey have led to increase data rates and densities for new generation multi-beam systems. It needs to process the data sets rapidly. However, some hydrographic softwares edit data using manual interactiveplot. The ratio of survey time, to the time involved in editing is about 1:1. Therefore, automatic approach must be developed. This dissertation combines median filter, local variance estimation and wavelet analysis to remove outliers and noises. Because median filter may distort details in datasets, so it is not considered as ultimate result, but is used to calculate local variance. Following, outliers are removed relying on threshold calculated by the local variance. After the datasets have been processed, however, some noises exist in it. Allowing for the advantage in removing noises, wavelet analysis is used to remove noises. The advantage of the algorithm is that can process large quantities of data, rapid and robust etc.Then, SSS image processing techniques are discusseed. Generally, sidescan image exists geometric distortion and intensity distortion. The former is discrepancy between the relative location of features and their true location, the latter is deviation from the ideal linear relation between image intensity and backscattering strength. If the raw data include the navigation information, the geometric distortions can be easily corrected. If without, the attitude of vehicle can also be inferred from the sonar image. Of course, the precise inferred from image can not be so high. Slant range correction is first step in image processing and auto interpretation. Most traditional sonar systems apply a slant range correction based on the assumption that the depth in across track is constant. When seafloor sharply changes, this assumption may cause a feature to be placed either too far or too near in the across-track direction. Some papers integrate swath bathymetric data with sonar imagery to correct this error, but if position error in SSS is big, it may not be a good method. So this dissertation presents a method based on sound curve tracing. This method improves pixel position accuracy in sonar image. Modem sonar system has high-resolution angle estimation, and provides a prerequisite for sound curve tracing. In order to correctly interpret the features in sonar image by computer, intensity correction is vital. Reed et al. correct it through calculate the ratio of average intensity for the total area to average for pre record line, but they don't remove the strong backscatter areas, so the high intensities of these areas will be added to these record lines. So operate will change the relative intensity ratio in these situations. Allowing for the advantage of wavelet locating sharp signal, this dissertation adopts to remove true sharp area. Intensity correction is more rational after the process is completed. A simulated data verified the method.And then, it discusses multi-beam sonar and side-scan sonar data co-registering and fusing. It presents a new algorithm for co-registration. It is that combines bathymetry data with intensity image, and obtains the feature points through the minimal angles of line segment, then decides the corresponding image points through the maximal correlate coefficient in searching space. Finally, the second order polynomial is used to the deformation model. It is shown that this algorithm can be used in the flat seafloor or the isotropic seabed. After image has been co-registered, Wavelet analysis is used to fuse the images.Following, it discusses the methods of feature extracting and classification in sonar image. Side scan sonar imaging is one of important source for seabed study. To be utilized in other field, such as ocean engineering, the image needs to be classified according to seabed materials, hithis paper, sonar image is classified according to BP neural network, and Genetic Algorithm is adopted in training network. In order to prevent BP network from trapping a local ninimum, Genetic Algorithm is adopted in train network. The feature vectors are average intensity, six statistics of texture and two dimensions of fractal. It considers not only the spatial correlation between different pixels, but also the terrain coarseness. The texture is denoted by the statistics of the co-occurrence matrix. Double Blanket algorithm is used to calculate dimension. Because a uniform fractal may not be sufficient to describe the seafloor terrain, two dimensions are calculated respectively by the upper blanket and the lower blanket However, in sonar image, fractal has directivity, i.e. there are different dimensions in different direction. Dimensions are different in acrosstrack and alongtrack, so the average of four directions is used to solve this problem. Finally, the real data verify the algorithm.