基于特征的超声信号分类检测方法

Feature based method for classifying and detecting ultrasonic signals

超声检测技术在工业和科研领域得到广泛的应用。在恶劣的条件下,超声信号受到混响的干扰。传统的匹配滤波方法在这种情况下检测性能不理想。该文提出一种基于特征的检测方法,对接收信号进行分类再进行检测。该方法基于模式识别来区分是否存在回波。先利用Wigner-Ville分布和双谱提取信号的特征,然后进行主成分分析降低特征的维数,降维的特征向量送入有监督学习的分类器。实验表明,与传统的匹配滤波方法相比,该方法在-5dB的信混比时具有较好的检测性能。

Ultrasonic detection is widely used in industry and research. However, the received signals are often strongly disturbed by reverberations and traditional matched filters （MF） do not effectively remove these. An approach based on signal features used here where the received signals are first classified before detection. The algorithm is basically a pattern recognition approach to distinguish the presence of a target. Signal features are extracted using the Wigner Ville distribution （WVD） and a bispeetrum. The dimensionality of the feature vector is reduced through principal component analysis with the reduced vector sent to a classifier using supervision and training. Tests show that the method has a better detection effect at the signal-to-reverberation ratio （SRR） of around 5 dB compared with traditional matched filters.