Benchmarking deep learning models for surface defect detection: a reproducible and statistically-rigorous approach
Problem
Comparing deep learning defect-detection models is unreliable because varying dataset partitions and stochastic training introduce variable outcomes.
Why it matters
Unreliable model comparison can lead to wrong choices about defect-detection systems in manufacturing.
What the AI did
Proposed a statistically rigorous, reproducible benchmarking methodology using ANOVA and Tukey's test, designed for small datasets, demonstrated on object detection models with the Northeastern University surface defect dataset.
Results
not stated in source
Impact
Provides a reliable framework for comparing defect-detection models, per the abstract.
Record
| Venue | Journal of Intelligent Manufacturing |
|---|---|
| Published | 2025 |
| Work type | article |
| Authors | Darío G. Lema, Lídia Sánchez-González, Rubén Usamentiaga, Francisco J. delaCalle |
| Citations | 13 (OpenAlex, at import time) |
| References | 40 |
| Model or method | not stated in source |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1007/s10845-025-02672-8 |
| Open access | hybrid |
| Keywords | deep learning, defect detection, benchmarking, ANOVA, object detection |
Importance rating
incremental / niche
Reproducible statistical benchmarking for defect-detection models; a real but narrow methodological advance for manufacturing QC, no scale stated.
More in Engineering & Materials
| # | Paper | Year | Citations | Importance |
|---|---|---|---|---|
| 1 | Scaling deep learning for materials discovery | 2023 | 1,471 |
8
|
| 2 | Application of Machine Learning in Material Synthesis and Property Prediction | 2023 | 165 |
7
|
| 3 | Machine learning-guided discovery of ionic polymer electrolytes for lithium metal batteries | 2023 | 115 |
7
|
| 4 | Machine Learning for Structural Health Monitoring of Aerospace Structures: A Review | 2025 | 44 |
7
|
| 5 | Power System Transient Stability Assessment Based on Machine Learning Algorithms and Grid Topology | 2023 | 42 |
7
|