As mentioned before, the feature fusion is an efficient step in scene understanding. We proposed to combine between the outputs of CNN algorithm, where the final features can effectively represent the scene images