DeepSight: Land Use and Land Cover Classification Using Satellite Images

Abstract

Remote Sensing data is constantly on the rise with launches of various satellites around the world, generating a huge amount of data. This data is raw i.e. it lacks semantics. Due to the lack of semantics, this data is untapped. To fully utilize this data, we propose a classification method based on deep learning, deployed as a web service- DeepSight. DeepSight uses a convolutional neural network- SpectrumNet to effectively classify land use and land cover. After training the model over 27,000 images, an accuracy of 96.3% is achieved for the testing set whereas the validation set gives us an accuracy of 95.1%. Thus, the results show a fairly high accuracy rate of classifying the multi-spectral images and this can be further used by multiple domains requiring Remote Sensing data semantics relating to land use and land cover.