Detection of porcelain insulator damages by image processing

Fractures and cracks on the insulator cause undesired energy interruptions and great economic losses. In order to prevent these, the fractures and cracks on the insulator surface must be detected in a short time. In this thesis, damage detection was made using the image of a broken or cracked insulator. Image processing methods were applied on the obtained image on the MATLAB program and defective insulators were determined by classification methods. Since it is aimed to detect the damage on the insulator by clarification, some operations must be applied on the image until the desired stage is reached. First of all, with low-level image processing steps, unnecessary noises on the image were cleaned according to the needs and the image was given clarity. Then, histogram equalization process was applied to equalize the color distribution on the image. After the histogram equalization, the image was converted to binary mode by thresholding at the most appropriate threshold value. Finally, morphological operations were applied to clear the unnecessary white spots and clarified the damaged parts, and the image was brought to the desired point and the damages on the insulator were clarified. A data set was created from the data obtained as a result of image processing, and image classification was performed on Python using support vector machines, artificial neural networks and deep learning methods, so that damaged and undamaged isolators were detected. Training and testing of classification methods were carried out with a total of 348 pieces of data. At the end of the study, classification methods were compared, and test accuracy rates of 66% from support vector machines, 60% from artificial neural networks and 74% from deep learning were obtained.