artificially inflate confidence; it is used to expose the model to operational variability and to reduce overfitting to a limited set of training images. 4.4 Evaluation metrics Segmentation quality is evaluated with intersection-over-union (IoU) and Dice coefficient. For a predicted domain P and a geologist-validated domain G, IoU measures overlap relative to the union of both areas, while Dice emphasizes agreement relative to the combined size of the predicted and reference regions. IoU = |P intersection G| / |P union G| (1) Dice = 2 |P intersection G| / (|P| + |G|) (2) Object detection is evaluated with mean average precision at an IoU threshold of 0.5 (mAP@0.5). Operational performance is evaluated separately through review acceptance rate, mapping-cycle time and the proportion of low-confidence areas correctly flagged for expert review. This separation between model metrics and operational metrics is important because a useful geological tool must be both mathematically accurate and operationally adoptable. Table 3 - Pilot performance indicators for GeoLithAI Metric Scope Pilot target Interpretation Mean IoU Lithological-domain segmentation >= 0.60 Minimum overlap expected for useful firstpass mapping. Dice coefficient Domain-boundary agreement >= 0.70 Measures consistency between prediction and expert polygons. mAP@0.5 Detection of mesh, water, paint, support and other objects >= 0.70 Measures detection quality for occlusions and artifacts. Review acceptance Mapped area accepted without major edits >= 60% Indicates whether outputs are practical for geologist review. Mapping cycle time Image-to-interpretation support time 30% reduction Assesses operational acceleration relative to manual-only workflow. 5. RESULTS FRAMEWORK The intended output of the pilot is not an autonomous geological map, but a structured interpretation package for expert validation. The package contains: (i) the original image, (ii) a lithological-domain segmentation layer, (iii) an occlusion and operational-object layer, (iv) confidence scores, and (v) editable polygons for incorporation into a 3D geological modelling environment. This structure preserves the geologist as the accountable interpreter while reducing repetitive digitization and improving traceability. The expected result is a faster first-pass mapping support layer that highlights visually coherent domains and areas requiring review. Low-confidence predictions are not hidden; they are treated as valuable information because they identify where additional field inspection, better
RkJQdWJsaXNoZXIy MTM0Mzk2