A method of ultrasonic signal feature extraction based on wavelet packet and principal component analysis

In order to effectively identify different types of ultrasonic defect signals, a method based on wavelet packet decomposition and principal component analysis was proposed to extract the characteristic quantities. Firstly, the multi-dimensional eigenvector set is composed of the energy coefficients of wavelet packet decomposition. Then the multidimensional feature vectors are reduced by PCA method to get the fusion feature quantity. Finally, BP neural network was used to classify and test different types of defect signals, and the obtained results were compared with the classification test results of characteristic quantities without PCA processing. The experimental results show that the feature extraction method can classify the defect types effectively, and the testing speed is obviously improved.