Leak-Free Cross-Validated Stacking with Per-Architecture Calibration for Sand-Boil Segmentation in Earthen Levees

2026-07-28

AI Summary

This work addresses evaluation leakage in sand boil segmentation caused by data scarcity—such as synthetic images derived from the test set and overlap between ensemble weight tuning and test samples—by proposing a leakage-free cross-validated stacking framework. Within a five-fold cross-validation scheme, each fold rigorously excludes synthetic samples originating from the same source as the test images and aggregates predictions from five independently calibrated encoder-decoder models via a pixel-level meta-learner trained exclusively on out-of-fold data. Key innovations include a leakage-free evaluation protocol, an architecture-aware temperature calibration mechanism, and a mask-conditioned synthesis method requiring no additional annotations. Experiments reveal that high inter-model error correlation (0.894) limits stacking gains; nevertheless, the method achieves a mean IoU of 0.718 on an independent test set (averaged over three random seeds), substantially outperforming the original SandBoilNet (0.608).

Abstract

Sand boils, points where water seeping beneath an earthen levee re-emerges at the surface, are early warnings of internal erosion, and deep segmentation networks are increasingly used to find them in inspection photographs. Annotated examples are scarce, and two common ways of working around that scarcity quietly inflate reported accuracy: tuning ensemble weights on the same images later used to score them, and training on synthetic images derived from the very photographs held out for testing. We present a sand-boil segmentation framework that closes both loopholes. Every synthetic image carries a pointer to its real parent, and a per-fold filter excludes any image whose parent is held out; five encoder-decoder backbones are trained under five-fold cross-validation, calibrated by one temperature scalar each, and combined by a per-pixel meta-learner fitted only on out-of-fold predictions. On the held-out test set the proposed Updated SandBoilNet reaches an intersection-over-union of 0.707 over three seeds, against 0.608 for the published original re-evaluated on the same split. Under the stacking protocol the calibrated stack reaches 0.681 against 0.694 for the strongest fold-averaged member, so it does not improve on the best single model; eight meta-learner families reproduce that outcome, which we trace to a mean pairwise error correlation of 0.894 among members. A synthetic pool filtered for label fidelity lifts the champion to 0.718 over three seeds against a 0.707 control. We also introduce a mask-conditioned synthesis route that makes the conditioning mask the label by construction, giving labelled training images at zero annotation cost.

Problem

• sand-boil segmentation
• data leakage
• cross-validation
• synthetic data
• earthen levees

Innovation

• leak-free cross-validation
• per-architecture calibration
• mask-conditioned synthesis
• sand-boil segmentation
• out-of-fold stacking