Research On Seabed Sediment Classification Based On Sonar Image Denoising And Enhancement

In the complex measurement environment,the sonar system is affected by various noises,resulting in poor quality of the original sonar image.In addition,due to the difficulty and high cost of sonar data acquisition,and the irregular image processing process,there are few images that can be used for sediment classification,so the seabed sediment classification is a small sample classification problem.Under the condition of small samples,machine learning and other classification methods are prone to over-fitting,which greatly affects the accuracy of seabed sediment classification.Therefore,in view of the low quality of the original sonar image and the insufficient performance of the classification algorithm under the condition of small samples,this paper studies the image processing and classification methods of the seabed sediment sonar.A Bilateral Shrink Filtering and Prewitt Sharpening method for sediment image processing is proposed to improve the sediment image quality.A classification algorithm based on Bag of Words(BOW)modified Support Vector Machines(SVM)is proposed.Compared with the traditional SVM classification algorithm,the classification accuracy of seabed sediment under the condition of small samples is improved.The main research contents of this paper are as follows:Firstly,to solve the problem of noise pollution and low quality of original sediment image,a denoising method of Bilateral Shrink Filter is proposed.This method uses pixel difference and spatial intensity difference to process edge feature and noise of sediment image respectively,and solves the problem that image blur caused by Wavelet Filtering and Median Filtering has great difference in denoising effect for different types of sediment.The denoising contrast experiment of sediment image is carried out by using Bilateral Shrink Filter,Wavelet Transform and improved Median Filter respectively.The experimental results show that Bilateral Shrink Filter has stronger denoising ability.Secondly,the first and second order sharpening and enhancement methods based on seabed sediment image are studied for the problems of blur degradation and edge feature weakening after denoising.Among them,the first-order sharpening adopts Robert,Sobel and Prewitt operators,which has the advantages of simple calculation and outstanding enhancement effect on mutation feature points.The second order sharpening adopts Laplace operator,which can detect more gradient feature points.Experimental results show that the first-order sharpening method based on the Prewitt operator has obvious advantages in enhancing image edge features and improving image contrast.Finally,the BOW-SVM classification method is proposed to solve the problem of low accuracy of seabed sediment classification under the condition of small samples.This method can extract texture features and convert them into semantic information,which can better distinguish edge features and texture structure in the sediment image,and improve the accuracy of the classification algorithm to recognize feature information.VGG,SVM and BOW-SVM methods are used to classify and compare the original sediment images and six processed sediment images respectively.The experimental results show that BOW-SVM improves the classification accuracy of small sample seabed sediment images compared with the classical SVM classification method.In addition,on the basis of denoising and enhancement processing of sonar images,the classification method is improved to obtain high-quality sonar images and solve the problem of low classification accuracy under the condition of small samples,so as to improve the classification accuracy of seabed sediment to a greater extent.