Research On Target Recognition Method Of Bathymetry Side-scan Sonar Image

At present,the long-distance detection technology for the seabed mainly uses sonar equipment,side scan sonar can measure detailed seabed details,and multi beam sounding sonar can measure the terrain information of the seabed.At present,most of the researches only deal with the sounding or side scan sonar data alone,but some special targets or terrain must combine the two kinds of images to recognize them well.The side scan sonar image is composed of the echo intensity of the seabed,but there is no corresponding height information.The bathymetric data can reflect the topography information of the seabed,but can not judge the geomorphic information.In view of the problem of recognition of seafloor mound,only relying on side scan sonar image can sometimes be confused with seafloor sediment,so the height information in sounding data can be introduced to distinguish.In order to solve the problem of false detection of seafloor mound,this paper studies the target fusion recognition method of side scan sonar image of bathymetry,at the same time,obtains the characteristics of seafloor topography and geomorphology,and improves the recognition accuracy.The main work of this paper is as follows:First of all,the preprocessing of the sounding sonar image and the side scan sonar image is studied.Aiming at the geometric distortion of the sonar image,the geometric correction method of the sonar image is studied,including the slant correction and velocity correction method of the side scan sonar image.In view of the problem that the sonar image obtained by the existing methods can not reflect the real topography of the sea bottom,this paper proposes a method to restore the sonar data to point cloud and then reduce the dimension,which can reflect the real topography of the sea bottom through the experimental verification.In addition,in order to facilitate the target recognition of the subsequent echo side scan sonar image,the registration method of the echo side scan sonar image is studied.Secondly,the segmentation of the sounding sonar image is studied.In view of the less characteristics of the sounding sonar image,this paper uses the traditional segmentation method to segment the sounding image.By comparing some traditional segmentation methods through experiments,Kmeans clustering algorithm is finally determined to segment the sounding sonar image,and the best performance is achieved in all indicators.Thirdly,the basic research of semantic segmentation model for side scan sonar image is introduced,the core structure is described in detail,and the data set building method and evaluation index for side scan sonar image are given.Due to the complexity of the side scan sonar image features,this paper uses Deeplabv3 + to do semantic segmentation of the side scan sonar image,and analyzes the shortcomings of the existing methods from the segmentation results and indicators.Finally,in order to improve the segmentation accuracy and incomplete segmentation,this paper proposes a semantic segmentation algorithm of side scan sonar image based on the improved Deeplabv3+.In order to improve the model and optimize the backbone network structure,a new normalized GN,activation function LRe LU and a new upsampling Dupsampling module are introduced.A large number of depth separable convolutions are used to obtain segmentation results step by step,which makes the segmentation more precise and reduces the loss of upsampling information and incomplete segmentation.The effectiveness of the algorithm in this paper is verified by comparative experiments Compared with the original network,the standard is 2.4% higher,which can be used for reference in the processing and analysis of semantic segmentation in side scan sonar image.Then aiming at the information complementarity of sounding side scan sonar image,this paper proposes a decision-making level based image fusion method of sounding side scan sonar segmentation,which is verified by experiments that this method can accurately identify the sea bottom mountain and sediment.Compared with the single sonar image target recognition method,this method has better detection performance and robustness,and provides an effective method for seabed target detection Solutions.