基于对抗残差网络的复杂海况舰船目标类型识别技术研究OA

In recent years,with the application of deep learning theory in the field of underwater acoustics,significant progress has been made in the research of underwater target recognition technology.However,in engineering application practice,the recognition model obtained by combining traditional feature extraction and classifier methods is difficult to maintain laboratory performance.The complex and variable ocean channels cause severe distortion of the acoustic signal before sensor reception,leading to feature mismatch and over-fitting problems in the recognition algorithm,and the algorithm performance drops sharply.The Adversarial Residual Neural Network(ARNN)model is proposed to solve the above issue.The model uses a Gradient Reversal Layer(GRL)structure and dual label adversarial training framework to compensate for the differences in channel propagation under different hydrological conditions,enabling the algorithm to focus on features that can characterize the essence of the target,with stronger robustness and higher recognition rate.To verify its effectiveness,two experiments are designed,using the mechanical radiation noise signals of ship targets collected multiple times in different sea areas and hydrological conditions in the South China Sea.Training and testing sample sets are created to train and test the algorithm model.The results show that compared to traditional network models such as Support Vector Machine(SVM),Convolutional Neural Network(CNN)and Residual Network(ResNet)model,the proposed ARNN model can effectively alleviate feature mismatch and over-fitting problems,making the model portable under different hydrological conditions and solving key problems in the application of artificial intelligence technology in water target recognition engineering.