Deep Learning for Fracture Detection in Borehole Images

A mid-sized Middle East carbonate operator turned 14 wells of manually picked borehole image logs into a DETR-based fracture and bed detection model that achieved 85% sensitivity at play scale - replacing per-well manual sinusoid picking with end-to-end transformer inference that predicts depth, dip, and azimuth in a single forward pass.

At a glance

Three metrics frame the shift from manual interpretation to transformer-based detection.

The challenge

Manual identification of fractures and bedding surfaces from borehole image logs - data from two different microresistivity imaging tools - sits at the bottleneck of every carbonate play characterisation workflow. Each sinusoid crossing an image unwrap represents a planar feature intersecting the wellbore; extracting its depth, dip, and azimuth requires an interpreter to pick the trace, fit a curve, and calculate geometry. Across hundreds of metres and dozens of wells, this is slow, subjective, and inconsistent.

The operator faced a mature Middle East carbonate play with 14 vertical wells and no scalable path to consistent dip and azimuth extraction at play scale. Existing supervised learning approaches - Mask R-CNN, U-Net segmentation - struggled with overlapping fractures in a single image patch and required mask creation as a preprocessing step. The team needed an end-to-end architecture that could handle multiple intersecting sinusoids, predict geometry directly, and generalise beyond the training wells without per-interpreter calibration.

What we did

We built a Detection Transformer (DETR) model with a ResNet-10 backbone and transformer encoder-decoder that processes borehole image patches and directly predicts sinusoid depth, dip, and azimuth - no mask creation, no post-processing curve fits, no heuristic fusion of overlapping detections.

The training corpus comprised 2,291 non-overlapping image patches for the fractures-only model, drawn from 14 vertical wells in the Middle East carbonate play, and 1,492 for the combined beds-and-fractures model, drawn from 11 of those wells. To overcome data scarcity, we applied augmentation via colour jitter, Gaussian noise, blur, and sharpness transforms - expanding the effective training set while preserving sinusoid geometry.

The DETR architecture handles multiple intersecting fractures in a single image patch by design: the transformer decoder attends to the full feature map and emits a set of object queries, each predicting a feature type with its depth, dip, and azimuth. Bipartite matching during training assigns each predicted query to a ground-truth sinusoid, eliminating the need for non-maximum suppression or hand-tuned overlap thresholds.

We trained separate fractures-only, beds-only, and combined models, measuring performance at multiple depth and angle thresholds to capture interpreter tolerance. At an 8 cm depth threshold - the distance within which a predicted sinusoid must fall to count as a true positive - the fracture model achieved approximately 85% sensitivity. Dip accuracy was about 90% at a 3° threshold, and azimuth accuracy reached about 93% at 20°.

Bed detection reached an F1 score of approximately 72% at a 9 cm depth threshold, with azimuth mean absolute error of around 20° and dip accuracy stabilising at about 94% beyond a 3° threshold. The lower F1 leaves room for improvement, but the geometric accuracy remains within interpreter tolerance for reservoir modelling.

The outcome

At an 8 cm depth threshold - the window within which interpreters consider a pick 'correct' - the fracture model delivered approximately 85% sensitivity, meaning it recalled 85 out of every 100 manually labelled fractures. Dip accuracy was about 90% at a 3° threshold, and azimuth accuracy reached about 93% at 20°, both well within the tolerance bands that feed reservoir models and stress field inversions.

Bed detection hit an F1 of about 72% at a 9 cm depth threshold, with azimuth mean absolute error around 20° and dip accuracy at about 94% beyond 3°. Bedding recall leaves room for improvement, but the geometric predictions remain fit for purpose.

Critically, the DETR architecture handled multiple intersecting fractures in a single image patch without mask preprocessing or heuristic fusion. Where Mask R-CNN required one fracture per crop and post-processing to merge overlapping detections, the transformer decoder emitted a set of sinusoid predictions in a single forward pass, each with depth, dip, and azimuth ready for geological interpretation.

What this unlocked

The immediate unlock was speed: sinusoid identification that takes days by hand runs in minutes, turning a days-long manual picking campaign into a validator-reviewed output in hours.

Consistency is the second unlock. Manual picking varies interpreter-to-interpreter, well-to-well, even shift-to-shift for the same person. The DETR model applies a uniform decision boundary trained on 14 wells of expert picks, eliminating the drift that complicates cross-well correlation and stress field inversion.

Scalability beyond the training play is not yet shown. On 5 horizontal wells, a different well geometry, the model reached about 55% at a 4 cm threshold, so each new field or well geometry needs its own labelled wells before the model is trusted there.

Lessons and next steps

Those thresholds are not arbitrary. An 8 cm depth window and 3° dip threshold reflect the geological uncertainty already baked into reservoir models, so clearing them at about 85% sensitivity and 90% dip accuracy is what separates a validator-ready output from a research-stage one.

Next steps focus on multi-well joint inference - treating a pad or field as a single graph where cross-well fracture correlations constrain per-well predictions - and active learning loops that flag low-confidence picks for human review, turning every correction into a training sample. The bottleneck in borehole image interpretation was never data volume; it was the assumption that a human must sit between every image and every insight. End-to-end transformer models break that assumption, turning sinusoid picking from a per-well manual task into a scalable, validator-ready output.

References

1 Carion et al. (2020). End-to-End Object Detection with Transformers (DETR). ECCV 2020. https://arxiv.org/abs/2005.12872

2 He et al. (2017). Mask R-CNN. ICCV 2017. https://arxiv.org/abs/1703.06870

3 Performance metrics (about 85% sensitivity at 8 cm, about 90% dip accuracy at 3°, about 93% azimuth accuracy at 20°, bedding F1 about 72% at 9 cm, about 55% at 4 cm on 5 horizontal wells) from the peer-reviewed fracture paper's review revisions and supplementary tables, on a 14-well Middle East carbonate dataset with 2,291 fracture patches.