A Study on Efficient Image Classification of Historical Monuments Using CNN

With the rise of tourism and data democratization, among many classification issues is the recognition of landmarks in the field of vision and perception, which is being actively researched. After spending so many years classifying structures and monuments in general from photographs, fine-grained challenges are now the subject of attention. This research paper presents an investigation into the classification of Indian monuments using a convolutional neural network (CNN), experimenting with ResNet50, InceptionResNetV2, EfficientNetB1, EfficientNetB3, and MobileNetV2. The experimented model was trained on a dataset of 24 different types of Indian monuments, consisting of over 4,000 images, with the goal of accurately classifying new images of Indian monuments based on their type. Various hyperparameters such as batch size, learning rate, and epochs were tuned to enhance the accuracy of the model. Additionally, data augmentation techniques, for instance random rotations, zooming, and flipping, were utilised on the training photos to increase the model's resilience. The trained model achieved an accuracy of 13% from the ResNet50 model to over 98% from the MobileNetV2 model (training accuracy), demonstrating the effectiveness of the final architecture and the tuned hyperparameters for the classification of Indian monuments.