Evaluation of hyperparameters in CNN for detecting patterns in images

Deep learning techniques have emerged as an effective solution to the problems of current pattern recognition techniques, such as neural networks. Within these new techniques, the convolutional neural networks (CNN) offer an integration to the recognition of patterns in images, given by the traditional set of images processing plus neuronal networks. This article presents the analysis of the different hyper parameters that imply the training of a CNN, which allows to validate the effects on the accuracy of the network. It is used as a base the recognition of electric energy meters, obtaining a network with an accuracy of 96.32 %.