Research On Classification Method Of Submarine Substrate Type Based On Characteristics Of Sonar Image

With the development of sonar technology,researchers have found that the images obtained through it,that contains rich information on the characteristics of the substrate,which can be used to understand the type of substrate and serve as a new auxiliary means for the underwater survey,inversion geomorphology and military operations.Based on the basic principles,characteristics and influencing factors of sonar imaging,this paper studies and analyzes various techniques of denoising enhancement,feature extraction and classification recognition.According to the advantages and disadvantages of the method and the simulation results,an appropriate processing method is selected to form a complete system based on the image processing and analysis of the submarine sonar.It has far-reaching significance and value for submarine research.Firstly,this paper briefly introduces the purpose and significance of the research on the image processing and classification of submarine sonar,analyzing the research status and progress of submarine sonar detection and image classification at home and abroad,explaining the process of sonar image generate based on the working principle of the detection system.It also briefly introduces the source of the sample data set used in this topic.Secondly,considering the main factors affecting the resolution of the sonar image of the submarine sediment and the source and nature of the noise,the multi-preprocessing methods applicable to the sonar image are analyzed and studied in a targeted manner.Through analysis and comparison of the simulation results,the sorting adaptive median filtering is selected,which performs noise reduction processing on it.At the same time,considering the low resolution and poor contrast of sonar image,the adaptive enhancement method based on Curvelet transform domain is adopted to deal with it,which has obvious advantages in improving texture detail and overall contrast.Thirdly,in order to improve the accuracy of image recognition classification,it is necessary to perform effective feature extraction processing before the classification.According to the unique edge,texture and statistical information of the submarine sonar image,the scale invariant feature transformation,gray level co-occurrence matrix and the improved gray-primary symbiosis matrix is studied and simulated in this paper.The first method is simple and convenient,and the extraction speed is faster.The feature matrix extracted by the second method can represent different bottom materials.In the third method proposed in this paper,the Canny edge extraction algorithm and the gray-primitive method are combined to combine the edge shape statistical features with the gray correlation to extract the feature information more accurately.The three methods start from many aspects,which is beneficial to match different types of classifiers to achieve better classification recognition.Finally,three classification algorithms,SVM,BP and CNN,are used to study the classification of submarine sonar images,and the suitable feature extraction method is matched in a targeted manner.Through analysis and comparison,CNN model has higher accuracy and is more suitable for submarine sonar image classification processing,so it is used as the final classification method.Through the dynamic adjustment of parameters,the classification accuracy rate is increased to 96%.Finally,tensorboard is used to display the model frame composition and accuracy curve,and the effect is good.It is verified that the CNN model has high application value and research significance in the classification and recognition of submarine sonar image.