Research In Composite Insulator’s Hydrophobicity Detection Based On Sparse Representation Algorithm

The composite insulator has been widely used in electrical power system due toits favorable hydrophobicity and hydrophobic migration characteristics. Duringoperation, the composite insulator’s hydrophobicity is decreased or even lost by theintegrated influence of ultraviolet light, pollution, electric field and so on. Theflashover voltage of composite insulator is thereby reduced, increasing the risk offlashover and affecting the stability of the power system. Therefore，it is urgent tocarry on researches on the detection technology of composite insulator’shydrophobicity. The existing testing and evaluation methods are still not mature, mostof them are relied on manual operation or complicated experimental steps. This paperintroduces a state-of-the-art classification method of sparse representation toautomatically classify the hydrophobicity level of composite insulators and hasdeveloped a set of application software system based on VC6.0platform, providingnew ideas for online detection of composite insulator’s hydrophobicity.In the automatic identification pattern, the hydrophobic image is improvedthrough enhancing methods based on Retinex model in preparation for the upcomingpattern recognition process. The image’s quality has been improved to some extentthrough global random path Retinex algorithm, but this method needs complicatedcalculation and could easily bring about irrelevant information. Local Retinexalgorithm including single-scale and multi-scale algorithm are therefore used toprocess gray image with high fidelity.Wavelet threshold de-noising method is used to de-noise and smooth theenhanced image. Soft threshold function, hard threshold function and semi-softsemi-hard threshold function which is improved from the soft and hard thresholdfunction are respectively applied. By comparing and analyzing the de-noising results,it is concluded that the semi-soft and semi-hard threshold function can preferablysuppress noise in the hydrophobic image and does not cause the edge of the imageblur.The sparse representation classification algorithm is used to automaticallyidentify hydrophobicity level. It uses the smallest norml method to calculate thecoefficient of sparse representation and searches the training sample image that bestmatches the test image by calculating the minimum residual image, therefore the testimage’s hydrophobic HC level is accurately identified. The algorithm has successfully avoided complicated feature extraction in general pattern recognition algorithms andhas certain robustness for different shooting conditions. The experimental resultsshow that the method can be effectively applied in the composite insulator’shydrophobic image classification.In VC6.0platform, the hydrophobic image acquisition software is developedusing DirectShow technology, and the hydrophobic image analysis software isdeveloped using OpenCV technology. These two software have realized the functionof hydrophobic image dynamic acquisition, manual and automatic identification ofcomposite insulators’ hydrophobicity level. The software have simple and clearinterface with stability.