Target Detection In Side Scan Sonar Image

Side scan sonar can effectively obtain seabed information and is widely used in marine geological survey,determining seabed geomorphic types,detecting seabed sediments and so on.However,the complex underwater environment often makes the sonar image have the characteristics of low contrast,unclear target contour and serious noise interference.This brings challenges to the efficiency and accuracy of target extraction in side scan sonar images.In order to solve this problem,on the basis of the characteristics of side scan sonar images,this paper mainly studies the processing methods of side scan sonar image preprocessing,target region detection,target image segmentation and feature extraction.Details are as follows:(1)The sonar image is preprocessed.Firstly,the moving average method and edge detection method are used to detect the bottom of the side scan sonar image data respectively,and then for the problems of low image contrast and uneven gray level,histogram equalization,gamma transform,classical filtering method,guided filtering,BM3 D and other methods are selected for experimental analysis and comparison.The results of edge detection method and histogram can be obtained After equalization and guided filtering,the sonar image is optimized.(2)Target detection for sonar images.Firstly,the principle and processing process of onedimensional CFAR detection are introduced.For images,one-dimensional detection is extended to two-dimensional CFAR detection.The effects of different detection window structures on target detection are analyzed.A detection method combining sorted CFAR detection and cross detection window is proposed to suppress background interference compared with other detection windows The effect is good.Finally,the detection results are further processed to locate the potential position of the target.(3)Image segmentation of the target image.Firstly,the traditional Markov random field segmentation method is studied.According to the characteristics of the target with highlight area and shadow area,the linear iterative super-pixel segmentation and adaptive intensity constraint are improved.Compared with the traditional segmentation method,the improved image segmentation method not only improves the accuracy of the target,shadow and shadow The accuracy of background resolution,and can better suppress the background interference in pre segmentation.(4)Feature extraction of segmented image.Starting from the basic theory of geometric moment,Hu moment invariant is proved,and the invariant moment based on Radon transform and Zernike transform are constructed.The extracted target results are tested on the above three invariant moments to prove the effectiveness of moment invariant feature.