Desenvolvimento de Metodologia Inteligente para Classificação de Tipos de Isoladores em Redes de Distribuição

The present study demonstrates an automated methodology for image classification of three types of insulators in Medium Voltage Distribution Networks. This components are known colloquially as Pin, Polymeric and Saia Baiana Isolators. The classification process occurs through the use of digital image processing techniques (DPI) and computational intelligence. This methodology can be characterized by the following steps: image segmentation, dimensional attributes extraction and characteristic parameters calculation. In addition, it was developed an Artificial Neural Network (ANN) to treat this information. Thus, the data obtained in the DPI stage was used for the training of the chosen ANN - a Multi-layered Perceptron Network. A comparative study was carried out to identify the optimized number of neurons for the Neural Network hidden layer, as well as to evaluate the ANN performance parameters for the classification process. At the end of the evaluation, the system obtained satisfactory results, achieving 99 % of efficiency in identifying the type of component present in the image.