Multibeam Water Column Target Extraction And Three-Dimensional Reconstruction

Multibeam echosounder system is widely equipped by research vessels worldwide because of its excellent bottom detection performance and working efficiency.With the development of underwater acoustics,sensor,computer,digital signal processing and other technologies,as well as the upgrading of related hardware,the multi-beam system has the function of collecting and storing the water column data,which extends its application range from the seafloor to the whole water column.As the source of depth and backscatter intensity,water column data has the ability to visualize underwater targets and 3D imaging.Target detection based on multi-beam system has shown great value in fishery analysis,hydrological survey,geological structure investigation,gas leakage assessment and other fields.In this paper,the automatic extraction and accurate expression of targets during water column detection are studied from three aspects:preprocessing of water column data,segmentation target of image,surface reconstruction of point clouds.The main work is as follows:1.According to the principle of water column data acquisition and imaging,the preprocessing methods are summarized and improved.Volume backscattering intensity is calculated by referring to the correction of seafloor backscattering intensity.The performance characteristics of target,seafloor and side lobes in different view images are illustrated with examples.In addition,the paper summarizes the existing noise processing methods for water column images,and discusses their relative merits.2.In order to solve the drawbacks in current method,which is time-consuming and unable to optimize parameters,an automatic target extraction method based on depth data is proposed.The algorithm takes region growing as the framework.Firstly,it searches the extreme points of depth or the maximum point of intensity to symbolize the region after analyzing the characteristics of depth with kernel density estimation.Secondly,the RSF model is used to ensure the segmentation accuracy,that is,the level set function is designated as a growing criterion.Finally,the depth fitted by polynomial is used to limit the growing,and the area of interest can be obtained.Experimental results show that this algorithm can precisely segment targets and its noise immunity is strong.3.Two different modeling strategies are proposed to meet the research requirements of static and dynamic targets.Dynamic targets use Splatting to construct and render the point clouds with different backscattering strength values layer by layer.The course data is interpolated by water column images to obtain relatively uniform point clouds.And then static targets are reconstructed by α-shape contour detection,triangular mesh construction and Loop subdivision.Experimental results show that the method can get models that accurately reflect the characteristics of underwater targets.