Scene Classification Using Deep Learning Technique

Abstract

In recent years, deep learning techniques have been of more focus due to its remarkable results, especially in computer vision. Computer vision involves processing and understanding image data which is well represented and processed by deep learning techniques. It has overcome the traditional methods of image classification where handcrafted features were used to classify images. With the advent of deep learning, researchers drew their focus on automated feature extraction for image classification. Deep learning is a promising approach to perform classification-oriented tasks. In this paper, we have focused on classification of classroom scenes using deep learning techniques by automatically extracting features and classifying the scenes into meaningful labels. This novel approach tends to ease the task of monitoring students in the classroom by automatically extracting information and analyzing it. At first, we introduce some of the traditional techniques and their limitations. We then introduce ResNet architecture of 152 layers pretrained with ImageNet data to accomplish our institute’s classroom scene classification. Further, we demonstrate the results of our classroom data classified using ResNet architecture pretrained with ImageNet dataset.