FTA-FTL: A Fine-Tuned Aggregation Federated Transfer Learning Scheme for Lithology Microscopic Image Classification

Keyvan RahimiZadeh, Ahmad Taheri, Jan Baumbach, Esmael Makarian, Abbas Dehghani, Bahman Ravaei, Bahman Javadi, Amin BeheshtiarXiv 2025 · DOI · arXiv

Introduction

Identifying rock types, or lithology, from microscopic images is a cornerstone of oil reservoir characterization. However, the development of robust Deep Learning (DL) models for this task is hindered by two primary challenges: data scarcity and data privacy. High-quality geological images are expensive to acquire, and companies are often reluctant to share proprietary data due to competitive concerns and regulatory requirements.

To address these hurdles, this paper proposes FTA-FTL, a Fine-Tuned Aggregation Federated Transfer Learning scheme. This approach allows multiple organizations to collaboratively train a high-accuracy classifier without sharing raw data, maintaining privacy while overcoming the limitations of small local datasets.

Methodology

The researchers developed the FTA-FTL scheme in two distinct phases. First, they performed a centralized model selection to identify the most effective architecture for geological image classification. After evaluating several state-of-the-art networks—including DenseNet169, Xception, and ResNet101—Inception-ResNet v2 was selected as the base model due to its superior performance when paired with a custom Fully Connected (FC) classifier.

The proposed model architecture utilizing an Inception-ResNet v2 base followed by a custom classifier for lithology identification.

The core innovation lies in the Fine-Tuned Aggregation (FTA) strategy. Unlike standard Federated Averaging (FedAvg), which simply averages model weights from different participants, FTA utilizes the Golden Section Search (GSS) algorithm. This mathematical optimization technique determines an optimal "fine ratio" ($\sigma$) for updating the global model, ensuring effective convergence even when data across participants is non-uniformly distributed (Non-IID).

Dataset and Preprocessing

The study utilized a specialized dataset derived from core samples of an oil carbonate reservoir in the Southern Persian Gulf. The dataset focuses on three primary lithologies: Argillaceous Limestone, Limestone, and Dolomite.

Lithological column showing sample depths and representative microscopic images for the three target classes.

To mitigate the issues of small and imbalanced data, the authors employed extensive data augmentation, including flipping, scaling, cropping, and rotation. This resulted in a balanced "mini-dataset" of 2,127 images (684 Argillaceous Limestone, 633 Limestone, and 810 Dolomite), providing a stable foundation for training both centralized and federated models.

Key Results and Insights

The FTA-FTL scheme demonstrated that the "privacy tax"—the expected drop in accuracy when moving from centralized to decentralized training—can be virtually eliminated.

Training Scenario Accuracy
Centralized (Inception-ResNet v2) 90.68%
Federated FTA-FTL (IID Data) 90.24%
Federated FTA-FTL (Non-IID Data) 89.83%

The results show that FTA-FTL consistently outperforms standard FedAvg by approximately 1-2%. By optimizing the aggregation process rather than relying on simple averaging, the scheme handles data inconsistencies more effectively. Furthermore, the use of Transfer Learning proved essential, allowing the model to leverage features from general image datasets (like ImageNet) to excel in specialized geological tasks.

Conclusion

FTA-FTL represents a significant step forward for collaborative machine learning in the energy sector. By combining the data-efficiency of Transfer Learning with a refined Federated Learning aggregation strategy, the authors achieved high-performance lithology classification while respecting data privacy. Future work will focus on scaling this approach to larger, more diverse global datasets and further optimizing communication overhead between federated participants.

Summary updated

Abstract

Lithology discrimination is a crucial activity in characterizing oil reservoirs, and processing lithology microscopic images is an essential technique for investigating fossils and minerals and geological assessment of shale oil exploration. In this way, Deep Learning (DL) technique is a powerful approach for building robust classifier models. However, there is still a considerable challenge to collect and produce a large dataset. Transfer-learning and data augmentation techniques have emerged as popular approaches to tackle this problem. Furthermore, due to different reasons, especially data privacy, individuals, organizations, and industry companies often are not willing to share their sensitive data and information. Federated Learning (FL) has emerged to train a highly accurate central model across multiple decentralized edge servers without transferring sensitive data, preserving sensitive data, and enhancing security. This study involves two phases; the first phase is to conduct Lithology microscopic image classification on a small dataset using transfer learning. In doing so, various pre-trained DL model architectures are comprehensively compared for the classification task. In the second phase, we formulated the classification task to a Federated Transfer Learning (FTL) scheme and proposed a Fine-Tuned Aggregation strategy for Federated Learning (FTA-FTL). In order to perform a comprehensive experimental study, several metrics such as accuracy, f1 score, precision, specificity, sensitivity (recall), and confusion matrix are taken into account. The results are in excellent agreement and confirm the efficiency of the proposed scheme, and show that the proposed FTA-FTL algorithm is capable enough to achieve approximately the same results obtained by the centralized implementation for Lithology microscopic images classification task.
BibTeX
@misc{rahimizadeh2025fta,
  author = {Keyvan RahimiZadeh and Ahmad Taheri and Jan Baumbach and Esmael Makarian and Abbas Dehghani and Bahman Ravaei and Bahman Javadi and Amin Beheshti},
  title = {{FTA-FTL: A Fine-Tuned Aggregation Federated Transfer Learning Scheme for Lithology Microscopic Image Classification}},
  year = {2025},
  eprint = {2501.03349},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2501.03349},
}
Figure