Key Techniques Of Imaging Processing And Sediment Classification Based On Acoustic Backscatter Data

The development of sonar technology provides advanced technical means for the monitoring of marine underwater environment and target detection.Nowadays,the sonar image generated from the acoustic backscatter data has been widely used to analyze and interpret the information status of the seabed surface.However,there exist in some problems in sonar image,such as low resolution,blurred target edge,low contrast and prominent speckle noise,which seriously interfere with the application of sonar image.Therefore,acoustic data acquisition and analysis processing are carried out for the underwater information of relevant sea areas.Taking the sonar image generated by backscatter data as the research object,several key technologies in sonar image processing are studied,such as image correction and enhancement,strip information fusion,sediment classification.These contents and achievements are expected to provide clear and high contrast images for the follow-up underwater investigation information and basic application data.The main work and contributions of this paper are as follows:(1)In this paper,the scientific value and development of acoustic backscatter data are summarized.Secondly,the factors in the process of data processing and the corresponding correction research are analyzed.The key technologies of sonar image processing are classified.In addition,the research status and existing problems of sonar image correction and enhancement,strip information fusion and sediment classification are introduced.This paper expounds the research objectives,research content and gives the overall technical route.(2)The process of acoustic backscatter data and the analysis of the influencing factors are described.Moreover,the imaging principle and data preprocessing contents between multi-beam sonar and side scan sonar are differentiated.These sonar preprocessing results will serve as the basis for the follow-up research on key technologies.(3)Sonar image correction and enhancement algorithm under low contrast and high noise background is studied.Due to the radiation distortion,the side scan sonar image has obvious gray imbalance and dark edge problem.A correction and enhancement technique is proposed,which is a combination of non-subsampled Shearlet transform(NSST),improved multi-scale Retinex and sparse dictionary learning.Experimental results show that: the proposed technique is able to effectively improve the image contrast and reduce the noise interference.The enhanced image describes the clear target contour and terrain texture information.In addition,the gray statistical image reflects the changing characteristics of grayscale value with continuous enhancement.The objective index values show the superior performance in maintaining the image contrast,clarity,average information content and filtering ability.(4)The image information fusion algorithm in the overlapping area of adjacent strips is performed.When the side scan sonar is used for underwater information detection,the information detection in the overlapping area of adjacent strips is incomplete or the information on one side is lost.This paper proposes a fusion technology for adjacent strip information,which combines non-subsampled Contourlet transform(NSCT),summodified Laplacian energy filtering and improved dual channel pulse coupled neural network model.The experimental results show that: this technique can effectively eliminate the pseudo contour phenomenon of blocky flow and suppress the interference of high noise signal.The fused image can effectively weaken the side shadow of the underwater target measured at different field of view angles,and show clear target contour and rich characteristics of seabed topography.Many objective index values are usually optimal,reflecting the information quality of image edge,clarity and overall similarity.(5)The algorithm of sediment classification in complex environment is studied.The multi beam sonar image has the characteristics of shadowing and high noise spots,and there is the phenomenon of false marking when selecting training samples.These problems affect the quality of sediment classification.Therefore,we propose a sediment classification method based on the acoustic backscatter image by combining a stacked denoising auto encoder(SDAE)and a modified extreme learning machine(MELM).Compared with other feature extraction methods and classifiers,the proposed method can effectively extract the deep feature information of sediments.It can avoid the interference of residual error and error information.The overall prediction accuracy of sediment classification is improved.And the boundary continuity of sediment classification is ensured.