Real-time Deep Learning-based Scene Recognition Model for Metaverse Applications

Scene recognition is a type of image recognitiontask essential to the success of Metaverse applications. Previousworks have focused on Scene recognition tasks for real-worldenvironments. However, it is crucial to establish scene recognitionmodels that can be applied in Metaverse applications. This paperapplied two Convolutional Neural Networks (CNN) models: Sim-pleNet and AlexNet, to automatically recognize Virtual scenes.The models are trained on the Scene15 dataset of real-worldscenes and tested on virtual-world scenes. The models achieveda test recognition accuracy of 50.96% and 78.08%, and a test timeof 50.00ms and 10.00ms respectively on 7 different categories ofvirtual scenes.Index Terms—CNN, deep learning, metaverse, scene recogni-tion, virtual reality