Land Cover Classification from Time Series Satellite Images

Abstract

The earth surface is continuously observed by satellites which are continuously orbiting around earth. This results in the generation of large satellite image data sets. This data is of increasing spatial resolution and temporal density. Land cover classification is an important application of satellite imagery. The results obtained from land cover classification is used to observe land use and changes such as resource planning, to detect deforestation, desertification and water scarcity. Land cover classification also helps in identifying the changes in the land cover. Most satellites along with high-end sensors provides pixel wise data. Due to high resolution images, the number of pixels which are generated are in millions even for the study of a small area. Therefore there is a need to balance runtime and accuracy for this task. In this paper, we did a comparative study of the models comparing Multi CNN model with traditional models to classify land cover from multi-spectral temporal Landsat-8 satellite data.