The Research On Feature Extraction Of Target Acoustic Signal And The Key Technology Under The Complicated Background

With the armaments'informatization, intelligentization and miniaturization gradually, target detection and recognition technologies have been the frontier of military intelligence system, and the key of target recognition is the feature extraction. Compared with the temperature, shape, spatial location, quality distribution and other objective feature, the acoustic signal made by target can spread around the obstacles, and it is not limited to visibility, weather and other natural conditions. So the acoustic signal is easily to be obtained by acoustic detector, and it has been an important feature of target recognition. The features of objective acoustic signal are mainly extracted by means of the analysis of time domain, frequency domain and time-frequency domain at present. However, the target subjects to various background noise and clutter interference on the battlefield environment, which leads to the conventional feature extraction methods can not extract the target features effectively and rapidly.According to the low SNR and short-term stability of target acoustic signals on the battlefield, the conventional feature extraction methods of objective aocustic signals, the researching status quo of power spectral estimation at home and abroad have been summarized in the dissertation. AR model power spectral estimation has been used to extract the feature of objective acoustic signal. However, the estimation of sinusoidal components'frequency using the Burg method will generate the splitting of spectral lines and the shift of peak-positions, the improved power spectral estimation has been proposed to extract the characteristics of objective acoustic signal, namely, modified Burg method combined with the Welch method. The power spectral estimation of single objective acoustic signal and that of objective acoustic signal in the presence of interference have been obtained through the improved method respectively. The features of acoustic signal have been extracted from the spectral shape, undulation degree of spectrum, characteristic frequencies and energy distribution, and the grey relation analysis has been used to recognize the target. Compared with traditional methods, the experimental results show that the improved spectral estimation method can extract the feature of objective acoustic signal accurately.