Recently, remote sensing image (RSI) classification becomes a difficult task, which aims to determine the images depending upon the entities find useful for various application areas such as urban planning, land resource management, disaster management, traffic surveillance, etc. The latest advances of the deep learning (DL) models have gained significant attention on RSI classification. Due to the familiarity of the convolution neural network (CNN) in the image processing communities, several CNN based approaches have been developed to classify the RSI. In this aspect, this study designs a novel deep learning based feature fusion model for RSI analysis, named DLFF-RSI technique. The proposed DLFF-RSI technique involves pre-processing to improve the quality of the RSI in three different ways namely Gaussian filtering based noise elimination, contrast enhancement, and data augmentation. Besides, two DL based models namely Inception v3 and Densely Connected Networks (DenseNet201) are employed for feature extraction process. Finally, two feature vectors are fused together to raise the overall performance of the proposed model. In order to demonstrate the improved performance of the DLFF-RSI technique, a wide-ranging of simulations take place on benchmark datasets and the results reported the improvements of the DLFF-RSI approach compared to other approaches.