Image-Based Lithology Classification Using Hybrid Deep Learning Architectures and Diffusion-Based Data Augmentation: Application to Brazilian Pre-Salt Carbonate Reservoirs
Bibliographic record
Abstract
Abstract Visual inspection remains widely used for lithology classification, but manual approaches are prone to delays, errors, and subjectivity. To address these challenges, we propose a deep-learning pipeline tailored for a complex, imbalanced dataset of core images from offshore pre-salt carbonate reservoirs in Brazil. Our methodology integrates diffusion-based data augmentation and a hybrid architecture combining DINOv2 with Central Difference Convolution (CDC). A novel loss function, merging Focal and Center Loss, is employed to improve performance on hard-to-classify samples and sharpen decision boundaries. Diffusion augmentation introduces controlled variability, while dual-transformer encoding captures rich image features. The experimental setup involved evaluating the model across eight dataset partitions, with balanced accuracy as the primary performance metric. Human volunteers achieved a mean balanced accuracy of 57%, highlighting the difficulty of visual lithology classification. State-of-the-art deep learning methods in the literature reached up to 63% balanced accuracy on this complex dataset. In contrast, our proposed approach consistently achieved 70% balanced accuracy, demonstrating a substantial improvement over both manual classification and existing automated techniques. This performance improvement is notable given the dataset's complexity, particularly in contrast to small gains typically seen at higher accuracy levels. In high-stakes offshore environments, where classification errors can lead to wellbore instability or blowouts, even modest accuracy gains offer substantial operational value. Additionally, top-2 accuracy reached 91.86%, providing reliable support for expert decision-making in real-time lithology classification.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".