Research Of Sonar Image Target Recognition Technology Based On Deep Learning

Underwater target recognition technology based on sonar images is widely used,and civil and military technologies have great demand for them.In the face of an increasingly complex international environment,especially in the South China Sea,countries are eyeing,developing sonar image target recognition technology,strengthening naval power construction,coordinating the use of various means to maintain and expand national maritime rights and interests and properly respond to maritime infringements.It is important to maintain China’s freedom of navigation and the safety of the marine passage.Therefore,the research starts from the imaging characteristics of sonar images,and studies the noise characteristics of sonar images from the two aspects of underwater noise field characteristics and sonar imaging principle,analyzes the sonar image noise composition,and traces the noise formation principle,which lays a theoretical foundation for the work of target recognition of sonar images.Secondly,the dataset augmentation method is used to augment the sonar image dataset,and the deep learning denoising technique is used to preprocess the sonar image set.Then the research combines the deep learning semantic segmentation technique and proposes a new multi-scale convolution kernel fusion method to segment and identify the target in the sonar image.Finally,this paper also studies how to apply this algorithm to sonar image video sequences,and mitigates the problem of target recognition delay caused by deep learning based algorithms in video sequences.The experimental results show that,the target recognition method based on deep learning based on deep learning solves the problems of low intelligence degree,poor robustness and low recognition accuracy of traditional sonar image target recognition algorithm,and plays an important role in the practical application of sonar image target recognition technology.