Monuments Recognition using Deep Learning VS Machine Learning

Abstract

Egyptian monuments have too many features and information to investigate. Working on monuments in Egypt, which is one of the most remarkable industry, would surely have a huge impact on the Egyptian economy and development. Any culture, in general, is unique when it comes to the different features of monuments, writing, and music. Examining techniques to recognize these monuments is not easy to be done especially concerning the history and stories behind each monument. Monument Recognition is a challenging process in the field of image recognition and classification. Many obstacles have to be tackled since many factors can affect the recognition method. The purpose of this paper is to recognize monuments in images using traditional and non-traditional machine learning techniques. VGG16 and Resnet50 have been used as examples of non-traditional machine learning techniques(deep learning), while KNN is used as a traditional one for a comparison between both techniques. The benchmark dataset used is Indian monuments images, due to the lack of datasets for the Egyptian monuments. We started by training our model, then we moved for the cross-validation where we used the k-folds technique with 5 folds and 25 epochs per fold. The ResNet50 achieved the highest performance measures in classifying unseen data. Its accuracy is 88%, precision 88%, recall 87%, F1-score 87% and the area under the ROC curve is 0.96.