In order to solve the problem of small underwater target detection and recognition，the methods of sonar image filtering，image segmentation and target extraction based on side-scan sonar are studied. Conventional filtering methods are difficult to effectively remove the noise in the image，resulting in degradation of image quality. In this paper，the non-local means filter algorithm and GPU acceleration method are used to obtain better processing results of sonar images，and to meet the real-time requirement of small underwater target detection. At the same time，the expansion algorithm and Canny edge detection algorithm are combined to realize the effective distinction between true and false underwater targets.
[4] DOBECK G J,HYLAND J C.Automated detection andclassification of sea mines in sonar imagery[C]//Detectionand Remediation Technologies for Mines and MinelikeTargets II.USA:Proceedings of the SPIE,1997.
[5] REED S,PETILLOT Y,BELL J.An automaticapproach to the detection and extraction of minefeatures in side scan sonar[J].IEEE Journal of OceanicEngineering,2003,28(1):90-105.
[6] WU T,XIA P,LIU X M,et al.TS-MRF sonar imagesegmentation based on the levels featureinformation[C]//Ninth International Symposium onMultispectral Image Processing and PatternRecognition(MIPPR2015).Enshi:SPIE,2015.
[12] LIN Y,LV F,ZHU S,et al.Large-scale imageclassification:fast feature extraction and SVMtraining[C]//IEEE Computer Society Conference onComputer Vision and Pattern Recognition.ColoradoSprings:IEEE,2011.
[13] LOUIS L W.Navigation and control of the Nereushybrid underwater vehicle for global ocean science to11000m depth:preliminary results[C]//OCEANS 2008.Quebec:IEEE,2010.
