Contamination Grades Measurement of Insulators Based on Image Color Feature Fusion

Abstract:

An insulator contamination grades measurement method based on feature level fusion of visible image information in red green blue (RGB) and hue saturation intensity (HSI) color spaces is proposed. Optimal entropic threshold (OET) segmentation algorithm is adopted to segment insulator surface. Features of RGB and HSI color spaces are calculated separately. Meanwhile, feature selection based on Fisher criterion is applied to obtain features which have the ability to represent the contamination grades efficiently. Kernel principal component analysis (KPCA) is adopted to carry out dimensionality reduction fusion of the combination of features and obtain three dimensional fused features. Probabilistic neural network (PNN) is used to identify the contamination grades. The experimental results indicate that the feature level fusion of image information based on KPCA has capability to characterize the contamination grades comprehensively. Compared with recognition using RGB or HSI features solely, the proposed method can obtain higher recognition accuracy and realize the contamination grades recognition effectively. A new method for the prevention of pollution flashover is presented.