Jasim et al. (2025) DDMSA-U-Net: A Lightweight Deep Learning Framework for Multi-Spectral Change Detection for Agricultural Land Use Monitoring

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Springer Link (Chiba Institute of Technology)
• Year: 2025
• Date: 2025-10-08
• Authors: Laith Jasim, P. Girish, Harshitha Deepanjali, Sarala D V, Sahana M P
• DOI: 10.1051/itmconf/20257901056/pdf

Research Groups

Not specified in the provided text.

Short Summary

This research proposes a novel, light deep learning architecture, Depthwise Dilated Multi-Spatial Attention U-Net (DDMSA-U-Net), to enhance the accuracy and efficiency of agricultural change detection using multi-temporal satellite imagery, achieving 91.6-96.6% overall accuracy and Kappa values above 0.85.

Objective

• To develop and validate a light deep learning architecture (DDMSA-U-Net) that improves the accuracy of change detection in agriculture monitoring, specifically addressing limitations in boundary accuracy, sensitivity to spatial scales, and background noise in heterogeneous landscapes.

Study Configuration

• Spatial Scale: Landscape to regional scale, focusing on heterogeneous agricultural landscapes, utilizing data from Landsat-8 and Sentinel-2 satellites.
• Temporal Scale: Multi-temporal, covering seasonal crop observations from 2021 to 2024.

Methodology and Data

• Models used: Depthwise Dilated Multi-Spatial Attention U-Net (DDMSA-U-Net), integrating depth-wise separable convolutions, dilated multiscale feature extraction, and multispatial attention mechanisms.
• Data sources: Multi-temporal satellite imagery from Landsat-8 and Sentinel-2; seasonal crop observations for validation.

Main Results

• The proposed DDMSA-U-Net model significantly improved classification and change detection performance compared to traditional methods.
• Achieved an overall accuracy ranging from 91.6% to 96.6%.
• Obtained Kappa values above 0.85 for all evaluated cases.
• The model demonstrated usefulness for observing agricultural transitions such as crop rotations, fallow periods, and urban encroachment.

Contributions

• Proposes a novel, light deep learning architecture (DDMSA-U-Net) specifically designed for enhanced change detection in agriculture monitoring.
• Integrates depth-wise separable convolutions, dilated multiscale feature extraction, and multispatial attention mechanisms to improve spatial discrimination and reduce computational costs.
• Addresses critical limitations of existing remote sensing change detection approaches, including boundary accuracy, sensitivity to spatial scales, and susceptibility to background noise.
• Validated the model's effectiveness with high accuracy and Kappa values using real-world seasonal crop observations over a multi-year period.

Funding

Not specified in the provided text.

Citation

@article{Jasim2025DDMSAUNet,
  author = {Jasim, Laith and Girish, P. and Deepanjali, Harshitha and V, Sarala D and P, Sahana M},
  title = {DDMSA-U-Net: A Lightweight Deep Learning Framework for Multi-Spectral Change Detection for Agricultural Land Use Monitoring},
  journal = {Springer Link (Chiba Institute of Technology)},
  year = {2025},
  doi = {10.1051/itmconf/20257901056/pdf},
  url = {https://doi.org/10.1051/itmconf/20257901056/pdf}
}