Research On Underwater Target Recognition Based On Deep Neural Network

Automatic recognition of underwater targets is the key technology of underwater acoustic equipments and intelligent weapons.All countries in the world attach great importance to related research work.Based on the practical problems of underwater passive recognition technology,the paper studies from three aspects:1.Signal pre-processing.Pre-processing includes denoise,pre-emphasis,framing and windowing.Environmental noise is one of the factors which have a great influence on the underwater acoustic signal.It will have a significant impact on the recognition performance.Therefore,the method of improving signal-to-noise ratio of signal is firstly studied.Taking into account the characteristics of the signal itself and the high frequency attenuation of the propagation process,the high frequency components are boosted by pre-emphasis to obtain robust feature.The basic idea of STFT transform,as the method of achieving signal analysis and feature extraction,is to make window processing of the signal in order to let a short signal be seen as a smooth signal;then analyze the results by adopting Fourier transform.In the meantime,a large amount of sample data can be obtained at the same time to avoid the problem that the training data is difficult to fit due to the insufficient number of samples.2.Feature extraction and multi feature fusion.How to extract the characteristics of the target signal,which can characterize the target signal itself,and distinguish between the other signals,is the key content of the research.The feature extraction of ship target radiation noise is carried out,and the extracted features imitate human auditory perception mechanism from different levels.The feature fusion is achieved by serial fusion and principal component analysis(PCA)is used to reduce the dimension,and the new features obtained have better classification performance.3.Classification and identification.The deep learning theory is introduced into the research of underwater target recognition,and a deep belief network is constructed,improved and optimized.Experiments show that the deep belief network constructed in this paper has a high classification accuracy for the target-radiated noise of the ship and can accurately predict the target category.At the same time,a good prediction result can still be obtained under the condition of low SNR,the generalization performance is superior to the traditional machine learning classification algorithm.