In order to solve reduction of pattern recognition accuracy when partial discharge ultra-high frequency（UHF）signal is distorted

a novel feature parameters extraction method based on textural features in time-frequency（TF）representation image is proposed. Firstly

the TF representation image of partial discharge UHF signal can be obtained by the s-transform. Secondly

the textural features of TF image can be extracted by the gray-level cooccurrence matrix（GLCM） method. Through the principal component analysis（PCA） method

the eigenvector consisted by textural features can be dimensionality reduced. The feature parameters and eigenvector of par-tial discharge UHF signal can be obtained and inputted into the support vector machine（SVM）classifier for pattern recognition. The test and recognition results show that the proposed feature parameters can be used to recognize 4 types of typical partial discharge UHF signal in transformer

and the highest recognition accuracy reached97.50%.