Identification of Deity Images Using CNN and Transfer Learning Models

The Indian subcontinent has a Hindu majority of 85%, Muslims as the minority of 9.1%, and other religious minorities of 5.9% who are the followers of Christian, Sikh, Buddhist, Jain, and animist religions. Each religion worships different Deities. This paper demonstrates the robustness of the deep learning model for the identification of different types of Indian Deity images with transfer learning models. Work is carried out using own dataset which comprises 11 categories of Indian Deity images, each of 200 images, contributing to total of 2200 images. Convolution Neural Network (CNN) is used to build the model with 3 transfer learning techniques: MobileNet, VGG16, and Xception for identification of Deity images and resulted in an accuracy of 98.76%, 96%, and 94%, respectively. The developed mobile app identifies the input Indian Deity image and displays brief history of identified Indian Deity (in both text and audio). In addition, the app also lists the five famous religious places in India with web link, contact details, details of the place, and location (Google map) of particular places related to identified Deity. Tourism is one the major source of income to improve the GDP of any nation. This app can be used to provide brief history of Indian Deity to school kids as well as, tourism app to promote tourism of religious places in India.