Research On Classification Algorithm For Submarine Sonar Image

With the continuous progress of sonar technology,Oceanologists use underwater sonar technology to probe the mysterious underwater world,The obtained sonar image of the submarine bottom contains a lot of subsurface information,which lays a foundation for the subsequent work of seafloor detection,inversion of the submarine topography and military operations.This subject is based on sonar imaging technology.Considering its imaging principle and interference factors,this paper analyzes several image denoising,enhancement,feature extraction and classification algorithms.Based on the experimental time and accuracy of each group,a group of algorithms that are most suitable for classifying submarine bottom sonar images are finally selected for comparison and analysis,which will have a certain driving effect on the research of bottom floor quality.First,a brief overview of the research background,purpose,and significance of submarine-quality sonar image processing and classification in this research direction,the research status and development history of this field at home and abroad are compared,and combined with the detection system workflow and its related principles.In addition,the source of the data samples used is introduced.Secondly,in view of the complex and changeable marine environment and the influence of external factors from more aspects of the equipment on the seabed bottom image,this article selects some preprocessing methods to reduce external interference.Through the analysis and comparison of several algorithms,sorting adaptive median filtering is used to denoise the original sonar image.In addition,it also processes the denoised image through the Retinex enhancement algorithm.The image resolution,contrast and other characteristics are improved to improve texture detail information and overall effect by it.In order to achieve better results in image recognition and classification in the later stage,a more appropriate and effective feature extraction algorithm must be applied to the image before processing the image.Taking into account the characteristics of the submarine sonar image itself,this paper uses three algorithms.They are: SIFT feature extraction algorithm,LBP feature extraction algorithm and DMD feature extraction algorithm.The SIFT feature extraction algorithm has certain uniqueness,efficiency,and scalability.It can accurately and quickly obtain the feature parameters in the sonar image.The scale of LBP feature extraction algorithm descriptor is relatively small,and the extraction is convenient and the calculation is simple.DMD feature extraction algorithm is a relatively new algorithm used in texture image classification.It is rich in content and has strong representativeness.This experiment applied it to submarine quality sonar images.Finally,the classification method uses SVM and DBN classification algorithms to perform classification verification on the descriptors obtained by the feature extraction algorithm mentioned above.Through the analysis of the experimental results and the process,it is found that the combination of DMD feature extraction algorithm and DBN classification algorithm has better performance and higher accuracy,and more suitable for submarine sonar image classification.This model was finally selected as the final classification method,and this experiment also provided a certain reference value for the study of submarine bottom image classification.