Cavitation Noise Classification Based on Spectral Statistic Features and PCA Algorithm

Small amount of training data confines the performance of auto noise classification system, especially when the dimensions of features are in a large scale. In this paper, 26 dimensional features are extracted from cavitation noise spectrum and line spectrum from three classes of cavitation noises. Principal component analysis (PCA) based method is applied to deal with the high-dimensional features which may lead to a high risk of over-fitting. Experiments using noise signals indicated that feature extracting method proposed in this paper performs well, and PCA processing is efficient to deal with the high-dimensional problem and can achieve a high recognition rate under the cases such as auto classification when the amount of training data is limited.