Research And Realization Of System Of Acoustic Target Recognition

With coming of a time that the automation, intellectuality and information play an important role, detection, recognition and locating of acoustic targets have been widely applied in civil and military aspects,such as detection and recognition for speech information and machine faults, and in warfare, automatic detection, recognition and locating for various vehicles, tank, helicopter, ship and submarine, etc. Due to the extensive application values of acoustic targets detection and recognition, the relevant system has become important study contents.Under background of detection and recognition of acoustic targets,passive acoustic recognition technique and system realization is proposed in this thesis. By the means of combining wavelet analysis, artificial neural network and Support Vector Machine(SVM), an acoustic targets recognition system has been realized. Initially in this paper, author presents a hardware system and an approach to restraining noise signal both with wavelet and wavelet packet algorithm. Simulation results show that wavelet packet method is better than the other on restoring original signal, and then a better reliability of the system have been achieved. A technique of energy feature extraction, in allusion to four targets data, has been carried out by wavelet algorithm. The results confirm that the feature extracted by this algorithm can express the characteristics of targets very well and obvious difference among various targets can be obtained. Finally, a BP neural network classifier (with input nodes number: 8, hidden nodes number: 7, output nodes number: 4) and a two-layer Support Vector Machine classifier that can identify three types of targets have been designed. Moreover, the analysis of advantages and disadvantages of two classifiers has been made.The experimental results about four targets show that the system designed by this study can effectively classify acoustic targets and the means suggested in this paper is feasible to acoustic target recognition.