Utilizing Deep Learning Techniques for Categorizing Architectural Heritage Images

In this research, we address the challenge of manually categorizing architectural heritage images, which is time-consuming and subjective. We propose an automated image classification method using Python and machine learning, specifically deep learning. The system recognizes various architectural features like altars, apses, bell towers, columns, domes, flying buttresses, gargoyles, stained glass, and vaults. Our approach employs a deep learning model trained on annotated images of buildings. Users can upload architectural photos, which are then scaled, converted into a numpy array, and fed into the deep neural network for prediction. The model predicts the architectural category and provides the likelihood of the prediction. The system displays the predicted category together with the actual image, prediction accuracy, and a description of the identified class. Additionally, a bar graph visualizes the class probabilities for better interpretation. The dataset includes 10 categories split into training and testing folders, comprising 5937 and 1434 images, respectively. The evaluation results indicate that DenseNet121 achieved 86.61% accuracy, ResNet50v2 reached 84.24%, and MobileNetv2 achieved 74%.