| With the development of maritime power strategy,maritime silk road strategy and anti-intervention/area denial strategy,as well as the progress of quiet target technology,China’s maritime power is gradually moving from shallow sea to deep sea,the traditional target identification method is difficult to adapt to the needs of target detection under the new situation,and it is urgent to develop new technology and equipment for active target identification in deep sea environment.The difficulties of deep-sea active target recognition are three: 1.the long-range target echo is weak,and the feature information available for recognition is small;2.the spatial and temporal fluctuations of the underwater background and the three-dimensional inhomogeneous distribution characteristics of deep-sea acoustic propagation lead to unstable target features and contamination by channel feature coupling;3.the actual measured well-marked data is difficult to obtain.To address the problem of difficult data acquisition,this paper explores the use of a typical simulation method that treats the formation of echoes as a linear system and simulates the received signal.Combined with the typical channel environment in the deep sea,the generic(Generic)model of Benchmark submarine and segmented(Segmented)model submarine are used as target objects,and the Bellhop ray acoustic model and COMSOL simulation software are used to model and simulate the typical channel environment and scattering characteristics in the deep sea respectively,and the background noise is simulated according to the Wenz deep sea noise spectrum The target signal library was established.Data pre-processing using short time Fourier transform,based on integrated learning method,three classifications of generic submarine,segmented submarine and ocean noise identification,to explore deep sea active target identification methods,as follows:(1)Underwater target scattering modelling and numerical simulation.The scattering of submarines under different directions of incidence was simulated using the boundary element method,and the difference between Benchmark segmented submarines and general-purpose submarines was compared to obtain the complex sound pressure at 1000 m of backscattering under different angles of incidence to form a spectrum,and the unit impulse response of target scattering was obtained using the inverse Fourier transform.The scattered impulse response is analysed using the short-time Fourier transform to compare the scattered wave characteristics of different submarine models and different angles.(2)Target echo modelling and pre-processing for deep sea channel coupling.Taking a location near the Bus Strait as an example,the topographic elevation dataset and the marine environment variable dataset are combined to simulate the channel unit impulse response under different scenarios using Bellhop.Assuming that the effect of scattered waves from other directions of the target can be neglected,a target echo model with typical channel coupling in the deep sea is established and over 200,000 echo signal data are formed by numerical calculations.Using short-time Fourier transform pre-processing,the training set and test set are designed to lay the foundation for integrated learning.(3)Target classification based on integrated learning.Based on the pre-processed dataset,recognition classification research is carried out to compare and analyse the performance of 1D Res Net-34 network,2D classical Res Net-18,Res Net-34,Res Net-50 and integrated learning network for target classification of deep-sea channel coupling,laying the theoretical and simulation foundation for further real-world data classification.The simulation environment in this paper is ideal and aims to provide an idea for similar target classification under channel interference for subsequent research. |
