Portal de Periódicos da CAPES

Artigo Acesso aberto Produção Nacional Revisado por pares

BIBTEX RIS

GPU Cuda JSEG Segmentation Algorithm associated with Deep Learning Classifier for Electrical Network Images Identification

2018; Elsevier BV; Volume: 126; Linguagem: Inglês

10.1016/j.procs.2018.07.290

1877-0509

Francisco Fambrini, Yuzo Iano, Diogo Gará Caetano, Abel A.D. Rodriguez, Clodoaldo Moya, Eduardo Carrara, Rangel Arthur, Frank C. Cabello, João Von Zubem, Luis Mariano Del Val Cura, João Batista Destro Filho, José Roberto Campos, José Hiroki Saito,

Advanced Measurement and Detection Methods

An automatic recognizer system based in Artificial Intelligence for thermographic images of the electric power distribution network is proposed in this article. The infrared thermography is usually used to conduct inspections in electrical power distribution lines, assisted by a human operator, which is usually responsible for operating all the equipment, selecting the hottest spots in the image (corresponding to the places needing maintenance), making reports and calling the technical team, which will do the repairs. The proposed automatic diagnosis system aims to replace the manual inspection operation using image processing algorithms. An old method of segmentation for thermal images known as JSEG is implemented and tested and a Deep Learning Neural Network is responsible to recognize the segmented elements. A comparison between the exclusive Deep Learning based image recognition with the same method preceded by the JSEG segmentation algorithm is done in this article, showing better performance with this previous segmentation of the thermographic images.