CNN Architecture For Land Classification Using Satellite Images

Abstract

Classifying land cover using satellite pictures is a crucial technique for researching terrestrial resources.Presently, several satellites, including Sentinel and Landsat 8, have captured enormous collections ofhigh-resolution images that contain satellite-based data. Due to the vast quantity of data and variety ofkinds, it is difficult to categorize the land cover in these images. Deep Neural Networks can categorizethese massive volumes of data, which makes them quite helpful in this situation. Similar studies in thefield, which need subject-matter expertise, relied on simpler models and a significant amount of manuallyconstructed parameters. In this study, a deeper Convolutional Neural Network (CNN) model without anysatellite imagespecific characteristics is proposed. On SAT4 and SAT6 pictures, our 10-layered networkexhibits exceptional accuracy of up to 96 per cent. It is still referred to as a lightweight model because themajority of models in artificial intelligence (AI)-CNN are far bigger and deeper than ours [1]. IndexTerms—sequential CNN, SAT4, SAT6, Convolutional Neural Networks, Remote Sensing, Satellite ImageClassification.