Identificação de fachadas residenciais e comerciais usando rede neural convolucional

Abstract

The classiĄcation of facades as residential or commercial is an important issue in several areas such as public safety, real estate analysis and urban planning. With Deep Learning (DL), convolutional neural networks (CNNs) are used for pattern recognition in images. In this work, a model based on the ResNet50 architecture is used, employing Transfer Learning to improve the performance in facade classiĄcation, adapting the pre-trained model to allow generalization to the speciĄc context of urban facades. In addition, Data Augmentation techniques are used to expand the set of available images and avoid overĄtting problems. The proposed approach uses Global Average Pooling layers to reduce the dimensionality of the data extracted by ResNet50 and dense layers for binary classiĄcation between residential and commercial facades. The loss function used is Binary Crossentropy, and the model is optimized with the Adam algorithm, which guarantees stability during training. The experiments carried out demonstrated that the combination of pre-trained models with a regionalized dataset allowed the creation of an eicient classiĄer for this task, achieving good levels of accuracy and area under the ROC curve (AUC). The results indicate that the methodology can be applied in automated systems for identifying urban facades, contributing to several areas, such as municipal management and the real estate sector