Research On Key Technology Of Side Scan Sonar Image Mosaic And Segmentation

Submarine geomorphy information is the key reference data of carrying out correlated activation in ocean,and side scan sonar is an efficient method to obtain it.There are several problems exist in original side-scan sonar image,such as larger geometric and radiation distortion which lead to low-quality of single image without geographic information.Currently,side-scan sonar image mosaic is mostly directly mosaic through geocode.Affected by towfish attitude and speed of vessel,this method can give rise to obvious image mosaic dislocation and deformation,it is hard to auquire entire high-quality image.Image segmentation is foundation and hingle to identify targets,existing segmentation methods are difficult to satisfy both efficienty and accuracy.In according to this,based on domestic and aboard researches,this article has carried on the following research based on original side-scan sonar data:1)Side-scan sonar data preprocessing has been carried out to improve quality of original side-scan sonar image and obtained a sonar image that truly reflects the Submarine geomorphy information.In order to decode from the original XTF data,the following aspects have been studied: submarine line extraction study,the slope can be obtained to correct the height data of towfish to the seabed;radiation distortion correction study to make sonar image more balanced and the target more clear;geometric distortion correction study that can precisely estimate geographic coordinates of scanning points.2)The side-scan sonar image mosaic based on the improved SURF algorithm is given,and the rate and accuracy of the mosaic are greatly improved.According to the characteristics of the blank area,the image is rotated and the blank area is deleted to achieve the purpose of accelerating the feature point detection;Set the common point slope and distance threshold to improve the point within the point ratio,the frequency and precision of the RANSAC algorithm are greatly improved,the fusion method of wavelet fusion and Laplacian pyramid fusion is analyzed quantitatively.It is found that the wavelet fusion effect is the best.3)The image segmentation based on the neutral set is studied,and the segmentation efficiency and precision are greatly improved.Starting from the neutral set,the gray level co-occurrence matrix of the image is constructed by the true subset of the neutral set and the uncertain subset,and the target fine texture can be achived;the experimental results show that the two-dimensional maximum entropy is more accurate than the maximum inter-class variance as the objective function of the segmentation threshold;The convergence and efficiency of the thresholds are compared with the particle swarm algorithm,the colony algorithm and the quantum particle swarm optimization algorithm.It is found that the quantum particle swarm optimization algorithm is optimal and the speed of the segmentation threshold acquisition is greatly improved;the shaded area can be quickly divided according to the gray histogram of the true subset.