Cross-regional land cover mapping from Sentinel-2 : machine learning vs. deep learning approaches

Name: Theeraya Uttha

Organization : Mahasarakham University. Faculty of Informatics

Email : theeraya.u@msu.ac.th

Name: Jantima Polpinij

Organization : Mahasarakham University. Faculty of Informatics

Email : jantima.p@msu.ac.th

Name: Samruan Wiangsamut

Organization : Mahasarakham University. Faculty of Informatics

Email : samruan.w@msu.ac.th

Name: Aniruth Chotthanom

Organization : Mahasarakham University. Faculty of Informatics

Email : anirut@msu.ac.th

Name: Thananchai Khaket

Organization : Mahasarakham University. Faculty of Informatics

Email : thananchai.k@msu.ac.th

Name: Khanista Namee

Organization : King Mongkut's University of Technology North Bangkok. Faculty of Industrial Technology and Management

Email : khanista.n@fitm.kmutnb.ac.th

keyword: Sentinel-2 (Satellite)

LCSH: Land cover

LCSH: Remote sensing

LCSH: Machine learning

LCSH: Deep learning (Machine learning)

Abstract: Land cover classification using satellite imagery is fundamental for environmental monitoring, agricultural planning, and urban development. This study investigates the performance of both classical machine learning models and deep learning approaches in a cross-regional setting using Sentinel-2 imagery. The EuroSAT dataset was employed for training, focusing on four land cover classes: Buildings, Water Bodies, Agricultural Lands, and Forests. To simulate real-world deployment, testing was conducted on a geographically distinct region in Kantarawichai District, Thailand. The study compares k-Nearest Neighbors (KNN) and Random Forest (RF) with sequence-based models including Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM) networks. Input data were preprocessed uniformly, with spectral bands normalized and represented as flat vectors for machine learning models and as ordered sequences for deep learning models. Evaluation metrics—Accuracy, F1-score, and Cohen’s Kappa— were used to assess performance. Results show that the LSTM model achieved the highest classification accuracy (87%), outperforming other methods in all metrics. The findings highlight the effectiveness of spectral sequence modeling in enhancing generalization to unseen regions and provide insights for scalable remote sensing applications.

King Mongkut's University of Technology North Bangkok. Central Library

Address: BANGKOK

Email: library@kmutnb.ac.th

Created: 2025

Modified: 2026-01-08

Issued: 2026-01-08

บทความ/Article

application/pdf

BibliograpyCitation : In IEEE Thailand Section. Antennas and Propagation/Electron Devices/Microwave Theory and Techniques Joint Chapter, Electrical Engineering/Electronics, Computer, Telecommunications, and Information Technology Association, and King Mongkut's University of Technology North Bangkok. The 5th Research, Invention and Innovation Congress (RI2C 2025) (pp.280-284). Bangkok : IEEE Thailand Section

eng

©copyrights King Mongkut's University of Technology North Bangkok

ลำดับที่.	ชื่อแฟ้มข้อมูล	ขนาดแฟ้มข้อมูล	จำนวนเข้าถึง	วัน-เวลาเข้าถึงล่าสุด
1	RI2C 2025pp.280-284.pdf	1.54 MB

ใช้เวลา

0.046645 วินาที