Classification of Indoor–Outdoor Scene Using Deep Learning Techniques

Scene classification is a process in which a computer’s visualizations of a scene are mapped to segments. Then, the machine applies deep learning to do the task. Indoor scene classification is more challenging than outside scene classification due to unpredictability. Over the last several years, numerous approaches for indoor scene classification have been created, and each of them is faced with a unique set of difficulties. Accuracy is the greatest issue with all of them. Since DL methods, in particular, CNNs, can automatically filter features without negatively impacting overall performance, they have become a practical solution for scene classification. CNN’s are a long-term technique used to classify images. Large-scale training dataset is needed to train for CNNs. Additionally, developing an entirely new CNN architecture is complicated from the bottom up. In this case, transfer learning, which provides desirable outcomes with small datasets, is the best approach. This work introduces a novel method for transferring images to the class using the CNN model and later also the VGG-19 pretrained model. The VGG-19 network is extraordinarily deep and was trained on a large number of different pictures including complex classification problems. Independently trained models are used to show the values through indoor and outdoor training programs. The experiment has been done on two different datasets like SUN397 and Places365 using both algorithms CNN and VGG-19. The comparison has been made of these algorithms into with existing AlexNet model and it is found that the VGG-19 has outperformed among all.