Improving Remote Sensing Classification with Ensemble Learning and XAI-based Interpretability

Radius Tanone (Universitas Kristen Satya Wacana)
Li-Hua Li (Chaoyang University of Technology)
Ramli Ahmad (Chaoyang University of Technology)
Alok Kumar Sharma (Chaoyang University of Technology)
Agus Cahyo Nugroho (Chaoyang University of Technology)

Jurnal RESTI (Rekayasa Sistem dan Teknologi Informasi)
Vol 10 No 3 (2026): Juni 2026

The challenge of classifying remote sensing images primarily stems from the unclear image quality produced by satellites during data collection. Relying on multiple models to recognize remote sensing images can sometimes lead to suboptimal performance. To address this issue, this study integrates transfer learning and ensemble learning to enhance the accuracy of single-model classification. This research employed pre-trained models, including EfficientNetB7, Vision Transformer, and ConvNeXt, and evaluated them on benchmark datasets RSI-CB256 and NWPU RESISC45. The results demonstrate that ensemble learning significantly boosts model performance beyond that of individual models. For the RSI-CB256 dataset, the average and geometric mean ensemble methods achieved the highest accuracy of 0.9985. For another dataset, namely NWPU RESISC45, the geometric mean ensemble achieved the best accuracy of 0.9720. Furthermore, this study explores model transparency and interpretability through eXplainable Artificial Intelligence (XAI) techniques. In addition to transparency, this study uses Grad-CAM to identify critical regions influencing the model’s predictions in remote sensing image classification tasks.