Scene classification involves grouping the images without semantic overlap which is an arduous task. There has been large amount of work carried out using CNNs in the literature. But, in the recent years Convolutional Neural Networks (CNNs) have been used in combination with other classifiers such as SVM to achieve better accuracies. In this juncture, we have proposed a novel architecture that combines CNN and Extreme Learning Machines for the classification of scene images. Also, we have proposed efficient feature extraction technique since ELM is sensitive to the number of features. Our experimental results show that the proposed method is able to produce state-of-art results on SUN (Scene UNderstanding)-dataset for scene classification. © 2019, World Academy of Research in Science and Engineering. All rights reserved.