PD pattern recognition using combined features

Abstract

For the purpose of identifying the defects within the insulation, a suitable set of combined features is used as input of back-propagation neural network (BPNN). In this procedure, fractal dimensions and the 2nd generalized dimensions of original PD images and fractal dimensions of high gray intensity PD images are proposed and computed by modified differential box-counting (MDBC) method, and thereafter moments and correlative statistical parameters are studied for recognition of PD images. Therefore feature vector consists of altogether 17 parameters. Meanwhile quadtree partitioning fractal image compression (QPFIC) is used for PD data compression in purpose of improving rate of PD image communication. With PD data gathered in artificial defect experiments, the final analysis results shows the method by means of combined features and BPNN performs effectively in recognition after QPFIC compression of PD images.