A Deep Learning Framework for the Classification of Indian Heritage Using Curated Datasets and CNN Models

The paper presents framework of image classification that uses deep learning techniques to classify monuments in India by using a very well-curated dataset that consists of 2,000 labeled images. This dataset consists of representations of landmarks that are significant, such as the Taj Mahal, Sun Temple Konark, Gateway of India, and Qutub Minar. Using TensorFlow and Keras libraries, the model was trained by an 80:20 train-test split, which used 1,600 trained dataset and 400 tested dataset of images. A simple CNN design was used, which had an few kind of convolutional layers with 32 and 64 filters, max pooling, and fully connected dense layers for classification. Data augmentation techniques such as ±20° rotatation, up to 20% zooming, and horizontal flipping were used to improve generalization, and images were altered so that pixel values were between 0 and 1. The model employed the Adam optimizer with a learning rate of 0.001 and categorical cross-entropy as its loss function. The model achieved a classification accuracy of 99.5% on the training data and 98.7% on the testing data after training for 25 epochs with a runset of 32 batches. From the loss and accuracy plots, it was seen that the model learned steadily and did not have significant overfitting. The following confusion matrix describes how predictions on new test images indicate very accurate identification of monument-specific features with only a small number of errors confirmed. The trained model can be applied practically to develop educational resources and preserve cultural heritage, and it is saved in HDF5 format. This work well shows how CNNs can recognize landmarks and can provide a strong foundation for further developments in monument classification.