Title: Land cover classification: a comparative analysis of clustering techniques using Sentinel-2 data

Authors: Mayuri Sharma; Chandan Jyoti Kumar; Aniruddha Deka

Addresses: Department of Computer Science and Engineering, Assam Royal Global University, Guwahati, Assam, India ' Department of Computer Science and IT, Cotton University, Guwahati, Assam, India ' Department of Computer Science and Engineering, Assam Royal Global University, Guwahati, Assam, India

Abstract: Automated land use land cover (LULC) classification may provide an authentic database of information to the policy makers in various fields like agro-climatic zone planning, waste land inventory projects, vegetation cover analysis, etc. It has a tremendous potential to contribute towards effective policy formulation. This article considers various unsupervised machine learning techniques: K-means, FCM, SOM, meanshift, GMM and HMM for land cover (LC) classification of Sentinel-2 data in the context of Assam, India. These models showed good performance in distinguishing vegetation cover area from the rest of the regions. K-means and FCM showed better performance in comparison to all the considered models. Meanshift correctly classified the continuous stretch of vegetation in study area 2. However, it misclassified between the vegetation and fallow land in study area 1. Similarly, in identifying built-up areas for study area 2 SOM covered the maximum but misclassified severely with other classes in study area 1.

Keywords: land use land cover; LULC; machine learning algorithm; K-means; fuzzy C-means; FCM; meanshift; self-organising map; SOM; Gaussian mixture model; GMM; hidden Markov model; HMM.

DOI: 10.1504/IJSAMI.2021.122008

International Journal of Sustainable Agricultural Management and Informatics, 2021 Vol.7 No.4, pp.321 - 342

Received: 12 May 2021
Accepted: 18 Jan 2022
Published online: 07 Apr 2022 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article