Feature recognition method based on fuzzy clustering analysis and its application

To deal with dimension disasters brought by high-dimensional data to intelligent learning algorithms, a feature recognition method by combining feature extraction with clustering analysis was proposed. The features of vibration signal were extracted by wavelet packet transform, and feature frequency bands of signal were identified and analyzed by fuzzy transitive closure method so as to realize further information compression. Two concepts as "compactness" and "degree of separation" between samples were proposed to determine the thresholds in fuzzy transitive closure method. A validity function was established to evaluate clustering models and determine the optimal clustering. Finally, the effectiveness and feasibility of the method were verified in the reduction of information dimension and feature recognition by a rotor test bed.