In this paper, we address the challenge of land use and land cover classification (LULC) using Convolution Neural Networks (CNN) on existing remote sensing datasets and compare the obtained results in an Indian Urban area context. This paper showcases the theoretical and experimental study of various large-scale, high-resolution remote sensing datasets. An image retrieval dataset is used to perform image classification and promising results are found. Also, a small-scale dataset is used and modified from a high-resolution large-scale data as per the requirement. Different datasets with different dimensions and spectral bands are used for the study. The results and comparisons of various datasets are tabulated. From a dataset point of view, classification or categorization techniques can be developed and assessed by making use of image retrieval datasets but this doesn't work the other way around.
This paper also provides the literature with standard outcomes for future research on datasets for Machine Learning based image classification especially in terms of reducing memory consumption of computers and fastening the process of execution. The resulting classification system finds its use in a large number of Earth observation applications.