Air Acoustic Target Identification Based On Neural Network Research

Use the sound wave of the airborne goal which produces in the flight process, to recognize the goal is the basic task of the passive sound detection system. The target identification technology belongs to the research scope of the pattern recognition, its key lies in the characteristic withdraw and the sorter design.The wavelet analysis is one brand-new signal processing method, which decomposes each kind of different frequency component to the frequency band that mutually not overlap. This method provides an effective way for the signal filter, the separation of signal and noise, and the characteristic withdrew. Therefore by earnestly studying in the wavelet analysis method, in view of the airborne target noise signal, the article adopts the wavelet soft threshold value to eliminate noises, obtains the satisfying effect.In view of the four kind of airborne targets , their noises characteristic was different or similar, this article proposed the characteristic withdraws by correlation method based on the time domain, and the characteristic withdraws by biggest entropy spectrum estimated which based on the frequency range. Aimed at the shortcoming that the characteristic dimension is big when using the biggest entropy spectrum estimate method, the article has applied the principal component analysis method to choose characteristic. The three methods have the high reliability to the airborne sound target, may as the sorter input.After the characteristic withdraws, this article applied three layers neural networks and the improved BP study algorithm to design the sorter and to carry on the study training under the different environment, finally produced the classified recognition experiment result.For the paper convenience, this article record the four kinds of airborne sounds goals as target A, target B, target C, target D respectively.