Pixel based classification for Landsat 8 OLI multispectral satellite images using deep learning neural network

Abstract

In satellite image analysis, classification of objects appearing in an image is a major step of information extraction. The information acquired from the satellite image can be used to provide the solution of various remote sensing problems such as agricultural challenges, natural hazards, and environmental monitoring etc. The most of the conventional classifiers used to classify satellite images do not provide realistic classification with high accuracy. Therefore, for achieving realistic and more accurate classification of the satellite images is still a challenging task. Machine intelligence is being harnessed to solve this task. In this paper, we present a novel machine intelligence pixel based approach for multispectral satellite image classification. We have proposed a machine intelligence deep learning neural network with five hidden layers and seven training features for satellite images classification. The model is tested on four types of data sets namely three classes, four classes, five classes and seven classes. Accuracy of 99.99%, 99.96%, 99.45% and 98.03% has been obtained on four test data sets respectively. In all cases, it is observed that the proposed method provides 8.52% higher accuracy than previously proposed classifiers.