Image-Based Iron Ore Feed Load Classification Using EfficientNet-B0 with Region-Density Attention and Hierarchical Feature Fusion

Iron ore estimation is a significant part of mineral exploration, as it supports determining both the quantity and quality of available mineral resources. Precise estimation facilitates effective mine planning, production management, and sustainable usage of natural resources. Conventional systems are mostly based on laboratory measurements and geological surveys that are frequently expensive, time-consuming, and restricted when handling larger-scale mineral datasets. In recent years, machine learning (ML) and deep learning (DL) systems have been progressively used to improve estimation by recognizing intricate patterns in mineralogical databases. This research introduces a Multi-modal Deep Learning Framework for Mineral Exploration and Iron Ore Load Prediction (MDL-MEOLP) model to enable precise mining decision-making. The proposed model performs hierarchical region generation utilizing a grid-driven patch extractor to capture multi-scale local regions. Subsequently, an EfficientNet-B0 backbone is employed for local feature extraction, while a region-density attention module emphasizes dense ore regions. Moreover, hierarchical feature fusion is applied to aggregate local-to-global representations and enhance global contextual understanding. Finally, a classification head with global average pooling, a dense layer, a dropout layer, and softmax activation is utilized to predict 17 feed load classes, and the model is trained utilizing Adam optimizer with cross-entropy loss over stratified data splits. To demonstrate the improved performance of the proposed MDL-MEOLP model, a comprehensive experimental analysis is carried out using Real-time Iron Ore Feed Load Estimation. The comparative results reported the effective performance of the MDL-MEOLP model with an accuracy of 97.58% over existing models.

@article{Karali2026,
  title     = {Image-Based Iron Ore Feed Load Classification Using EfficientNet-B0 with Region-Density Attention and Hierarchical Feature Fusion},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {17},
  number    = {7},
  year      = {2026},
  publisher = {The Science and Information Organization},
  author    = {Sami F Karali and Fawaz Alanazi and Eman Ramadan Elsharkawy and Asma A. Alhashmi and Abed Saif Ahmed Alghawli and Imen B. Mohamed and Chams Sallami and Abdulbasit A. Darem},
  doi       = {10.14569/IJACSA.2026.0170716},
  url       = {https://doi.org/10.14569/IJACSA.2026.0170716}
}