Automated Hailpad Dent Detection and Segmentation Using Machine Learning
Bibliographic record
Abstract
Hailpads are a widely used method for recording hailstone impact data by capturing indentations made by hailstones during hailstorms. As a simple and low-cost approach, hailpads are often deployed in vast networks to collect spatially-distributed data. However, the subsequent process of analyzing these indentations is often long and intensive (taking up to several hours for a single hailpad), and subjective. This research presents a novel, automated approach to identify the major/minor axes and depth distributions of hailpad dents via an image processing and machine learning pipeline, aimed at reducing the time and effort required to analyze the constituent dents of a hailpad. Using high-precision 3D scans, hailpads are depth mapped and then binarized based on user-prescribed adaptive thresholding, contrast equalization, and area filtering parameters. Next, the resulting binary masks are used as input to a convolutional neural network (CNN), which separates dents in clustered and non-clustered regions via instance segmentation. Built on a training dataset of simulated hailpad binary masks, the model was evaluated with a 93.0% Intersection over Union (IoU) score on predicted dent masks. In further comparisons against manual analyses and third-party commercial 3D scan assessments, the model excels in identifying individual impacts from within densely grouped regions. However, the overall prediction distributions are hindered to varying degrees by the influence of false positives in dent detection from non-hail artifacts in the input binary masks. Overall, this automated approach demonstrates the potential to considerably expedite hailstone dent identification and lays the groundwork for extracting more physical properties in the future, such as approximations for volume, impact velocity, and accumulated impact energy.
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 | 0.000 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
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".