RECOGNITION OF IMAGE PATTERNS OF OILS, BY CHARACTERIZATION OF COLOR SPACES WITH NEURONAL AND BAYESIAN CLASSIFICATION

In this article, we present a methodology that allows from color change, as a basis for characterization and neural networks and Bayesian algorithms as classifiers, different types of commercial oil to determine the level of wear. For the implementation of the state of the oil in-situ, an analysis was made for samples of seven types of oils, each of which was taken thirty images. To these images, the HSV transformation and the YCbCr transformation were implemented, and for each component, the value of mean quadratic root (RMS), standard deviation (STD) and the average value were extracted. The use of biplot allowed to determine the contribution of each feature to each main component which supported the decision to retrain the classifier with less redundant classes. At the end, a 100% classification is obtained both by Bayesian classification for the seven classes and by a neuronal classifier made up of two hidden layers, each of seven neurons, three input neurons and seven variables or output classes. The use of in-situ detection may generate diagnostic and maintenance tools in operating equipment