Room Surface Material Detection using Deep Learning to Determine the Reverberation Time of Classroom

Abstract

In this study, we propose a novel approach for determining the reverberation time of a classroom using image processing and deep learning. A pre-trained ResNet-50 deep convolutional neural network was fine-tuned on a dataset of images captured in various classrooms. The images were augmented and processed to extract relevant features, such as room dimensions and material properties, that were used as input to the network. The network was trained to predict the reverberation time based on these features. The results showed that the proposed approach achieved high accuracy and outperformed traditional methods for determining the reverberation time. This approach can be used in various real-world applications, such as acoustic design and optimization of classrooms for enhanced learning environments. From the model that has been developed, the image used is able to be classified according to their respective class, resulting in 98.0% accuracy, and the calculation of reverberation time of the classroom using the Sabine equation is able to conduct, with a result of 0.4823s.