Research On Seabed Sediment Feature Extraction And Classification Based On Side Scan Sonar Image

As one of the important information sources reflecting the terrain of the seafloor,the underwater sonar image has great guiding significance and research value in the field of deep sea exploration.At present,the image recognition technology is paid more and more attention.Extracting the features of the target area from the image to achieve the target segmentation,feature enhancement,feature classification and target recognition have been applied to so many aspects of real life and production.Obtaining image information and establishing an objective description of multidimensional images have far-reaching significance and value.Based on the comprehensive understanding of contralateral sonar imaging principle and image features,considering the submarine environment and the noise characteristics,this paper presents a complete image processing method based on denoising and enhancement,feature extraction,and classification recognition.Reveal the differences of characteristics and classification criteria of multi-type seabed quality data,and provide theoretical methods and technical support for the accurate identification and correct classification of seabed sediment information.First of all,the purpose and significance of seabed sediment sonar image feature extraction and classification are expounded.The current situation and progress of the research on the feature extraction and classification of seabed sediment sonar image feature both domestically and abroad are analyzed.The principle analysis and methodological introduction are given in terms of image enhancement,noise reduction and feature extraction.Combining with the characteristics of the side-scan sonar image,we give a targeted image processing method.Secondly,image denoising and enhanced preprocessing are performed.The adaptive median filtering,Gaussian filter and wavelet transform sample images are respectively selected for noise smoothing.At the same time,considering the low resolution of sonar images and the difference of contrast,the histogram equalization and linear segment enhancement are respectively used to adjust the gray level and local features of the denoised images.Images are better visualized and the features are clearer after preprocessing.Thirdly,according to the features of edge,texture and statistical information of submarine ground-truth sonar images,the gray-level co-occurrence matrix of image gray distribution,the point-covered fractal dimension model describing gray irregularity and the wavelet reflecting energy distribution The energy model is used to extract the multi-dimensional feature parameters of the image.A multi-level and multi-dimensional vector is constructed to describe the image,the image information is more accurately extracted.Finally,we choose SVM and AdaBoost classification algorithm to classify the seabed sediments.Firstly,the principles and learning strategies of the two classification algorithms are described in detail.Then the classification model is constructed according to the classification algorithm flow and the extracted feature vectors are input into the classifier to compare the classification results of the two classification strategies.The experimental results show that using SVM to classify the seabed sediment has higher accuracy.