Partial discharge image recognition using a new group of features

This paper presents a new group of features used for partial discharge (PD) pattern recognition, based on the description of detail and statistical characteristics of PD images by using fractal features and statistical parameters, respectively. An improved differential box-counting method is proposed for fractal dimension estimation of PD images. The new group of features is used as the input parameters of a back-propagation neural network (BPNN) for PD image recognition. During defect model experiments in the laboratory, five types of artificial defect models are used to acquire the data samples, which are used to qualify the proposed PD recognition method. Analysis results show that the proposed features are effective for PD images recognition