Scene Classification with Deep Neural Nets Using Background Suppression

Abstract

Deep Learning can be a powerful tool to replace the human eyes in the field of Scene classification. In this paper, we propose scene classification for four different classes using background suppression and Convolutional Neural Network. Background suppression is achieved with luma transforms. Supervised learning is used and the processed images are fed to the four layered convolutional neural network, which would enable the system to classify the images. Background suppression seems to indicate an increase in the specificity which is very much essential for scene classification. However, with increase in layers, sensitivity of the system does increase resulting in lesser validation loss. The training and validation accuracies have shown much improvement in comparison with the convolutional neural network and several other approaches.