Research Of Acoustic Target Recognition Based On Data Fusion

The acoustic target recognition is not only the key technology in target detection system, but also hot in the research fields of military guidance, security monitoring, intelligent transport and so on. It has an important research value because data fusion technology is used in target recognition, which can achieve the purpose of improving recognition success rate and recognition credibility.This thesis, which takes two types of vehicles as study objects, has investigated the technology of acoustic target recognition technology based on data fusion, system practicality and recognition performance, and mainly includes target detection, feature extraction, feature fusion and decision fusion.In the aspect of object detection and feature extraction, the interception of effective data segment is implemented by double threshold combined by short-term energy and zero crossing rates, and detection algorithm based on linear prediction coefficient distance, and then, the characteristics of the acoustic signal is extracted by AR parameter model and wavelet energy respectively. At last, test verification is executed based on theoretical analysis.In the aspect of data fusion, in view of the same kind of sensors, fusion recognition method of feature level and decision level is focused on to deploy investigation. On the feature layer, eigenvector is fused by BP neural networks; on the decision layer, fusion problem of target recognition is studied by D-S theory of evidence. By various of simulation experiments, data fusion of feature level is better than that of decision level from recognition effect. Fusion method of decision level based on D-S theory of evidence is less than that of feature level based on BP neural networks from the angle of calculation. Both of them have correct recognition rates over than single information source in the condition of higher Signal to Noise Ratio, which improved that the fusion system has more excellent performances.In the aspect of algorithm study, acoustic target recognition system based on DSP(TMS320F2812) which is consist of data acquisition module, serial-port communication interface, data buffer module and DSP is built, TMS320F2812 transplantation of each algorithm of the recognition system is realized, effective classification recognition of tracked vehicles, wheeled vehicles and background noise, the real-time property and efficiency of the system is verified.