Mashraqi et al. (2025) Hybrid deep learning and optimization-based land use and land cover classification for advancing sustainable agriculture in Najran city, Saudi Arabia

Identification

• Journal: Scientific Reports
• Year: 2025
• Date: 2025-11-25
• Authors: Aisha M. Mashraqi, Eman A. Alshari, Hanan T. Halawani, Ebrahim Mohammed Senan, Yousef Asiri, Bander Mohamd Alowadhi
• DOI: 10.1038/s41598-025-25908-2

Research Groups

• Department of Computer Science, College of Computer Science and Information Systems, Najran University, Najran 61441, Saudi Arabia
• Department of Artificial Intelligence, Faculty of Computer Science and Information Technology, Al-Razi University, Sana’a, Yemen
• Computer Science and Information Technology Department, Thamar University, Dhamar, Yemen
• Department of Computer Science, College of Applied Sciences, Hajjah University, Hajjah, Yemen
• Technical College of Telecommunications and Information, Jeddah University, Jeddah, Saudi Arabia

Short Summary

This paper develops and validates a hybrid deep learning model, integrating Convolutional Neural Networks (CNNs), Ant Colony Optimization (ACO), and Random Forest (RF), for accurate land-use/land-cover (LULC) classification in Najran, Saudi Arabia, using 2023 Landsat-8 imagery, achieving up to 97.56% overall accuracy to advance sustainable agriculture.

Objective

• To develop and validate an enhanced hybrid deep learning model for LULC classification using Landsat-8 imagery over Najran, Saudi Arabia, addressing the lack of optimization-enhanced deep learning architectures for heterogeneous environments. This aims to improve classification accuracy, minimize computational inefficiency, and enhance decision-making for sustainable land management.

Study Configuration

• Spatial Scale: Najran City, Najran province, Saudi Arabia (17.565604° N, 44.228944° E), covering an area of approximately 1156.15 square kilometers.
• Temporal Scale: 2023 (Landsat-8 satellite imagery).

Methodology and Data

• Models used:
  • Deep Learning CNN Architectures: LeNet-5, AlexNet, VGG19, GoogleNet, ResNet152, DenseNet121, MobileNetV2, ShuffleNet, SqueezeNet, RegNet.
  • Feature Optimization: Ant Colony Optimization (ACO) algorithm.
  • Classifier: Random Forest (RF).
• Data sources:
  • Satellite Imagery: Landsat-8 (2023) multispectral images (bands 1-7) with 30 meters spatial resolution, obtained from the United States Geological Survey (USGS) and NASA.
  • Ancillary Data: Normalized Difference Vegetation Index (NDVI) data (USGS or European Space Agency), SOI toposheet (1:50000 scale).
  • Software: SAGA GIS (v. 2.3.2) within QGIS (v. 3.6).

Main Results

• The hybrid CNN-RF framework achieved high accuracy for LULC classification in Najran City.
• The top-performing models were:
  • VGG19-RF: 97.56% overall accuracy, 0.972608 Kappa Coefficient, 98.92% Precision, 98.91% Recall, 98.91% F1-Score.
  • GoogleNet-RF: 96.15% overall accuracy, 0.983502 Kappa Coefficient, 95.83% Precision, 96.47% Recall, 96.15% F1-Score.
  • DenseNet121-RF: 92.39% overall accuracy, 0.973301 Kappa Coefficient, 96.78% Precision, 96.71% Recall, 96.72% F1-Score.
  • ResNet152-RF: 92.26% overall accuracy, 0.914909 Kappa Coefficient, 97.66% Precision, 97.61% Recall, 97.62% F1-Score.
• Class-based area statistics for Najran City indicated:
  • Built-up area: approximately 29%–33%.
  • Vegetation area: approximately 14%–25%.
  • Bare ground: approximately 9%–22%.
  • Water area: approximately 9%–22%.
• Ant Colony Optimization (ACO) for feature selection consistently outperformed Particle Swarm Optimization (PSO), with PSO-optimized models showing a 5.2% to 7.1% reduction in accuracy across architectures.

Contributions

• Introduces a novel three-layer hybrid model combining CNN-based feature extraction, Random Forest classification, and Ant Colony Optimization for LULC mapping, particularly in the Najran region.
• Develops a unified model that effectively captures spectral and spatial dependencies while minimizing noise and computational cost.
• Provides a comprehensive comparative evaluation of 10 hybrid CNN architectures, offering insights into how architectural diversity and optimization strategies impact classification performance in semi-arid environments.
• Offers a repeatable analytical framework for extensive LULC charting and observation, providing precise, time-consistent land-cover maps crucial for agricultural sustainability, urban growth management, and resource distribution.
• Supports national environmental goals and policies, including Saudi Vision 2030 and the Green Saudi Initiative, by enabling accurate land management.

Funding

• Deanship of Graduate Studies and Scientific Research at Najran University, Kingdom of Saudi Arabia, through the Nama’a program.
• Project code: NU/GP/SERC/13/383-4.

Citation

@article{Mashraqi2025Hybrid,
  author = {Mashraqi, Aisha M. and Alshari, Eman A. and Halawani, Hanan T. and Senan, Ebrahim Mohammed and Asiri, Yousef and Alowadhi, Bander Mohamd},
  title = {Hybrid deep learning and optimization-based land use and land cover classification for advancing sustainable agriculture in Najran city, Saudi Arabia},
  journal = {Scientific Reports},
  year = {2025},
  doi = {10.1038/s41598-025-25908-2},
  url = {https://doi.org/10.1038/s41598-025-25908-2}
}