Research On Detection And Recognition Of Acoustic Beacon Signal Based On Machine Learning

A black box is a specific data logger located on an aircraft,that emits periodic pulse signals of a fixed frequency after falling into the water.Therefore,we can search a crashed plane in the ocean by its emitting period pulse signal.Salvaging the wrecked aircraft involves the recognition of specific acoustic beacon signals,which is a big challenge in the time-varying,space-varying and frequency-varying marine environments.First,space-time joint processing based on synthetic aperture beamforming is performed to improve the signal-to-noise ratio(SNR)of received signals.Since the target acoustic beacon signal is a periodic pulse signal,two passive synthetic aperture algorithms are analyzed to derive the expression of the synthetic aperture beam power of periodic pulse signals.Numerical simulation results show that the extended towed array measurement(Extended Towed Array Measurements,ETAM)algorithm can improve the array gain and realize weak signal enhancement.Secondly,time-frequency analysis is combined with Support Vector Machine(SVM)for signal recognition.On the basis of adaptive line spectrum enhancement,five features in time domain and frequency domain are extracted after time-frequency analysis.A feature evaluation mechanism is established,and effective features are selected as the inputs of the SVM classifier according to the robustness and correlation criteria.The results of numerical simulation and atlake experimental data have verified the effectiveness of the proposed SVM-based acoustic beacon signal recognition method.Finally,considering the long pulse repetition interval(PRI)of the acoustic beacon signal,which results in a small duty cycle,research focuses on combination of the attention mechanism and neural network for signal recognition.Based on the Long-Short Term Memory(LSTM)neural network,the attention mechanism makes key frames in a signal a greater weight through keyvalue matching,and thus the neural network pays more attention to the periodicity of the signal.The results of numerical simulation and at-lake experimental data processing have shown that the recurrent neural network integrated with attention mechanism can effectively identify black box signals.