Edge Extraction Method Study On Seafloor Geomorphology Based On Image Analysis And Processing

The research and development of the ocean have become a hot spot.The seafloor geomorphology provides a basis for studying the movement of the continental plate and the occurrence of natural disasters on the seafloor,and has a guiding significance for the installation and laying of offshore platforms and submarine pipelines.So the seafloor geomorphology identification accurately and comprehensively is a fundamental project.This dissertation analyzes the features of depth image and curvature image from multi-beam data.Based on image processing algorithms,this dissertation studies and improves the method of extracting the seafloor geomorphology boundary.The proposed method mainly includes image preprocessing,image segmentation,and boundary extraction.First,this dissertation preprocesses the image,including graying,image filtering and image enhancement.In image enhancement algorithms,the homomorphic filtering algorithm,the histogram equalization algorithm and the image enhancement based on the Logistic retarded growth model are compared and studied.Then,the fixed threshold segmentation and adaptive threshold segmentation methods are studied.A global threshold segmentation algorithm based on pixel difference values is used to segment the depth image.A global threshold is calculated based on pixel statistics to segment the curvature image.The image segmentation is also implemented by using multiple thresholds.These thresholds are estimated based on histogram and local minimum values of the image projection.The Niblack algorithm,Sauvola algorithm,T.Romen Singh adaptive threshold algorithm with integral image are compared and analyzed.Then the local deviation calculation method in T.Romen Singh algorithm is improved by introducing the mean difference between a pixel and its neighborhoods.The improved algorithm enhances the quality of segmentation and saves the running time.Finally,the dissertation studies the boundary line extraction method,including boundary refinement,non-maximal suppression,single breakpoint complement,boundary length calculation based on 8-neighborhood boundary tracking,and short boundary filtering.This method precisely locates the position of the boundary line,realizes the refinement and extraction of the boundary line,further improves the accuracy of the boundary extraction,preserves the main boundary,and filters out the shorter boundary and noise generated during the measurement.This dissertation discusses the proper value of the neighborhood window size and the correction coefficient in the adaptive segmentation algorithm through experiments.The experimental results from different segmentation parameters show that the proper segmentation parameter is smaller for a smaller image area in the global single-threshold segmentation algorithm to obtain a more completed boundary.In addition,the experimental results show that in the image segmentation section,for the curvature image,the global threshold calculated from pixel statistics and the improved adaptive threshold methods can extract a better boundary of the geomorphology unit.The noise is greatly suppressed in the proposed method.The proposed boundary extraction process runs for less than 3 hours.