A new method of classification with rejection applied to building images recognition based on Transfer Learning

The present paper1 proposes a new method of classification with rejection for the scenario of building images recognition based on the probability vector generated by the classifier. Also, it is performed an evaluation of a set of pre-trained models of convolutional neural networks (CNN). Transfer Learning technique is used for features extraction (feature vectors), these are used to feed the classifier. Similarly, an evaluation is conducted on a set of classifiers with the objective of identifying the most optimal machine learning algorithm during the scene of buildings images recognition. The experiments are evaluated on the first version of the Cusco Building Image Dataset (CuscoBID). Finally, it is developed the second version of CuscoBID, composed of 4560 images of 14 different historical buildings, available to the entire scientific community.