| Artillery sound type recognition is the key technology of modern artillery reconnaissance systems.It aims to distinguish the type information of enemy artillery sound source by detecting and analysing enemy artillery sound signal,which is of great significance in battlefield information confrontation and other scenes.Due to the influence of many random factors such as atmospheric attenuation,multipath propagation,atmospheric stratification and flow,the mainstream artillery sound type recognition methods based on based on classical pulse waveform features are not adaptive to the long-range propagation distortion of artillery sound in different propagation distances and propagation environments,limited by the high requirements of the above characteristics for waveform purity.Therefore,it is of great theoretical significance and practical value to explore an artillery sound type recognition method that is different from the existing machine learning principles and more universal in application scenarios.In this paper,the methods of long-distance artillery sound type recognition under unknown conditions are studied.Two methods,based on machine learning and signal processing,are adopted to solve the adaptability to unknown propagation distance and unknown propagation conditions under long-distance propagation conditions.The main work and innovations are as follows:.(1)Aiming at the defects of the current artillery sound type recognition method based on the classical waveform characteristics of pulse signal,an artillery sound type recognition method based on LSTM neural network time-frequency feature fusion is proposed.Firstly,the time-frequency domain feature extraction is carried out for the continuous framing signal of artillery sound,and the time-frequency feature sequence with time continuity is constructed.Then,the time-frequency comprehensive information and continuity information of the timefrequency feature sequence are further fused and extracted by using LSTM neural network,so as to try to obtain the characteristics of artillery sound signal with stronger characterization ability and higher stability,thus to improve the adaptability and generalization ability of artillery acoustic type recognition technology to the distortion of long-distance propagation of acoustic signal.(2)From the perspective of signal analysis and processing,an artillery sound type recognition algorithm based on spatial projection of waveform signal is proposed.The algorithm is mainly based on the hypothesis that the long-range observation signal shares the same signal space with the short-range muzzle wave produced by the artillery of corresponding type,and the recognition of artillery sound type in unknown distance and environment is realized by the matching search maximizing the waveform space projection energy.To prove that the above tow methods are more adaptive to the acoustic signal distortion of long-range artillery under different propagation conditions,this paper uses simulation and field measured data to verify and analyze the performance of this algorithm in many aspects,and compares it with the existing methods based on the classical waveform characteristics of pulse signal.The results show that the above methods have better adaptability under long-range propagation conditions,It provides a potential and feasible scheme and idea to improve the sound type recognition ability of long-range artillery under any unknown distance and unknown environment. |
