Research On Classification And Recognition Method Of Surface Ship Noise Signal

The classification and identification of hydroacoustic targets is of great significance to China’s defence and security construction and the development of marine resources.Among them,effective ship target identification technology is the key to law enforcement and monitoring of infringing ships in sensitive waters.In order to achieve ship target recognition,feature extraction and classification of ship radiated noise is required.Aiming at the problems of small signal-to-noise ratio,insufficient feature characterisation capability,small number of samples and a wide range of recognition types that commonly exist in the current ship sonar signal recognition,this paper deals with them from the following aspects.Firstly,pre-processing of the ship radiation noise signal.The sound in the dataset used in the article contains human activity sound and marine mammal noise in addition to ship radiation noise,so audio screening,signal noise reduction,signal interception,signal enhancement and other processing are required.Secondly,finite number feature extraction is applied to classify and identify the ship radiation noise signals.Based on the Ensemble Empirical Mode Decomposition(EEMD),the Intrinsic Mode Function(IMF)obtained after the decomposition is investigated,and the selected feature extraction parameters are Mean Standardized Accumulated Modes(MSAM)principle high and low frequency energy difference method and sensitive IMF variance mean method.Thirdly,after applying a limited number of features for classification and recognition,problems such as the need for human-designed features,complex feature extraction process,low efficiency and consumption of large amount of human and material resources arise.Therefore,the study of neural network classification methods based on deep learning focuses on three types of convolutional neural network structures,namely Le Net5 neural network,improved Le Net5 neural network and Res Ne Xt neural network,introducing their basic principles and algorithm implementation in detail,giving the corresponding parameters and deriving the applicable occasions for each type of method.It is clear from the comparative analysis that Res Ne Xt neural network has the highest recognition rate and is more suitable when the network layers are deeper and more complex;the improved Le Net5 neural network structure has slightly higher performance than Le Net5 neural network and can be used when the database is simpler.Finally,on the basis of type recognition,we carry out the individual recognition of the same type of ships and different types of ships,and the experimental results further verify the practicability of the feature extraction method and the classifier model adopted in this paper.From the practical point of view,this research has great application value,which can not only realize maritime ship detection,port target classification management and biological reserve construction,in the civilian field,but also realize marine security maintenance,underwater target detection and identification,port area security and early warning.