Manuscript received October 6, 2025; revised November 17, 2025; accepted December 17, 2025; published October 19, 2026.

Abstract—This study investigates the application of DenseNet-121 for forest fire detection using cross-sensor transfer learning between Landsat-8 and Sentinel-2 satellite imagery in the Philippines. The primary objectives are to evaluate the impact of a locally fine-tuned Sentinel-2 dataset on model performance and to explore the model’s capacity for cross-sensor transfer learning from Landsat-8. DenseNet-121 was selected for its dense connectivity architecture, which enables efficient feature propagation, and was further enhanced using Gradient-weighted Class Activation Mapping (Grad-CAM) for spatial interpretability. The model achieved an F1-Score of 89.47% on unseen Sentinel-2 data, demonstrating strong generalization performance. Grad-CAM visualizations provided interpretable heatmaps for approximate fire localization and severity estimation, classifying alerts from low to extreme. These findings underscore the model’s potential as a scalable and explainable solution for forest fire detection, supporting timely decision-making in environmental monitoring and disaster response.

Keywords—forest fire detection, cross-sensor transfer learning, DenseNet-121, convolutional neural network, remote sensing, Landsat-8, Sentinel-2, satellite imagery
 

Cite: John Paul Q. Tomas, Lance Asher M. Elloso, Estefanee A. N. Garcia, Khristine F. Mecija, John Christian S. Diamante, Tiong Yew Tang, and Bonifacio T. Doma, Jr., "Forest Fire Detection: Cross-sensor Transfer Learning with DenseNet-121 from Landsat-8 to Sentinel-2 Satellite Imagery in Benguet, Philippines," Journal of Image and Graphics, Vol. 14, No. 5, pp. 859-867, 2026.

Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).