Research On Sonar Image Segementation Method Based On Wavelet Packet

The ocean is an enormous treasure of resources,and imaging sonar has been developed rapidly at the background of exploration for marine resources.Sonar image is a valuable tool for analyzing the sea bottom.It is necessary to analyze sonar image especially when work in the deep seabed area,such as detecting shipwrecks,exploring archaeological sites,detecting underwater mine,inspecting pipeline and other important tasks.Sonar image segmentation is an imperative step to analyze the image information,because the correct segmentation result is the precondition for extracting information accurately.However,due to the influence of suspended matter in ocean and the influence of uneven bottom quality,as well as the unstable change of acoustic channel,sonar image generally has a lot of noise.Therefore,in order to obtain the correct segmentation results,a filtering method suitable for sonar image is needed which can smooth the background noise while not destroying the edge information as much as possible.Wavelet transform and wavelet packet transform have the characteristics of multi-scale and multi-resolution,which can peel the image layer by layer,extract and process the information in the part of interest,so they are very suitable for image analysis.This paper mainly focuses on the research of sonar image filtering processing method based on wavelet transform and the research of sonar image segmentation method based on wavelet packet transform.The main work are as follows:On the one hand,it is about the filtering research of sonar image.In this paper,the types of sonar images noise were analyzed,among which speckle noise and environmental noise are the main ones.In order to solve the two types of noise,an improved Block-Matching and 3D filtering(BM3D)method based on wavelet transform is proposed.Through wavelet transform,can filter out part of environmental noise by abandoning the high frequency diagonal component.Then,exponential transformation and logarithmic transformation are performed on the reconstructed sonar image.The speckle noise of sonar image is converted into the Gaussian additive noise which is sensitive to the BM3 D method.Finally,the BM3 D method has the advantage of edge preserving and noise reduction,and the smoothing effect of the background area is achieved to the best.Experimental results verify the advantages of proposed method compared to median filtering,Kuan filtering,wavelet soft threshold filtering and traditional BM3 D.On the other hand,it is about the segmentation method research of sonar image.The requirement of sonar image segmentation is to effectively divide the target bright region,target dark region and seabed reverberation region of sonar image.The segmentation of sonar image in this paper is based on the edge of the region,and the correct segmentation of sonar image can be achieved by accurately detecting the edge of the target region.A improved edge detection method based on wavelet packet transform is proposed to solve the problems of poor connectivity,poor anti-noise performance and excessive false edges in current sonar image processing.After wavelet packet decomposition,the edge information of sonar image presents different characteristics in each sub-band,accordingly,different methods are used to detect the edge of the region in the high frequency component and the low frequency component.The low frequency component decomposed by wavelet packet is detected by morphological edge detection method in four directions.In the high-frequency components decomposed by the wavelet packet,Otsu method is used to distinguish edges and noise reasonably;Finally,the high frequency detection edge and the low frequency detection edge are reasonably fused to obtain the edge detection result,which can obtain continuous edge and influence of background noise is low,so as to realize the effective segmentation of the sonar image into three regions.Experimental results verify the advantages of proposed method compared to Otsu segmentation,FCM segmentation,Canny segmentation and wavelet modulus maximum segmentation.