Scene classification is basic problem in robotics and computer vision application. In Scene classification focused on complete view or event that contains both low and high level features. The main purpose of scene classification is to diminish the semantic gap in between social life & computer system. The main issue in scene classification is recognizing tall buildings, mountain, open country and inside city. We applied combination algorithms of feature extraction on trained datasets. Our proposed algorithm is hybrid combination of SIFT+ HOG named as HFCNN. As compare with the existing CNN architecture, HFCNN perform betters with high accuracy rate. Accuracy rate for proposed architecture is more than 96% as calculated with better time consumption and cost effective.