Eficiência de redes neurais convolucionais na identificação de edifícios em condições desafiadoras de imagens

This research focuses on the effectiveness of CNNs in identifying buildings in challenging images. The objective is to compare and improve model performance using CNNs. The methodology included image preprocessing, application of convolutional and dense layers with ReLU and Softmax activation functions, and training with cross-validation. The results indicated that the integration of algorithms can significantly enhance the ability of neural networks to handle variations in angle, lighting conditions, and obstructions in images. CNNs are effective tools in recognizing visual patterns in buildings, highlighting their potential to improve georeferencing practices and other applications.