Integrity Review Report: "Welding Defect Detection Algorithm Based on Feature Extraction and Extreme Value Search"

Academic fraud report · Geng Detector

Summary

This report reviews the article "Welding Defect Detection Algorithm Based on Feature Extraction and Extreme Value Search" by Liang Wei, Tao Liang, Zhang Guangxian, and Li Zhenhua, published in the Journal of Shandong University (Engineering Science) in 2014 (DOI: 10.6040/j.issn.1672-3961.0.2014.013). The overall verdict is questionable (yellow flag). The review is constrained by the fact that only the extracted text was available; no high-resolution figures or numerical data tables were provided, precluding pixel-level image analysis or statistical forensic tests. The principal textual finding is an internal inconsistency in Section 2.1 (Page 49): Equation (2) defines median filtering (G(i,j)=med[...]), yet the following prose describes replacing the center pixel with the mean of template pixels, which is the principle of mean filtering. This indicates careless copy-paste or textual splicing. Additionally, the abstract and conclusion claim "high detection accuracy" without providing any quantitative metrics (accuracy, false-positive rate, ROC, or comparison tables). Equipment references (Paxscan 2520, GE CCO 2505) appear temporally consistent for 2014. Image-related and data fabrication checks could not be performed. Confidence is moderate; findings are limited to text-level inconsistencies.

Verdict

Questionable (Yellow) — The paper contains a clear internal logical inconsistency between Equation (2) and its accompanying prose, and the claimed high detection accuracy is unsupported by any quantitative experimental results. Because only text was available, image-level and statistical data-level forensics could not be performed.

Key findings

Methodological self-contradiction (Section 2.1, Page 49): Equation (2) defines median filtering (G(i,j) = med [...]), but the following paragraph describes replacing the center pixel with the *mean* of the template pixels (the principle of mean/average filtering). This is a copy-paste or composition error that suggests hasty drafting.
Unsupported performance claims: The abstract and Section 4 (Conclusion) state that the algorithm offers "high detection accuracy" and "low complexity," but the entire paper contains no quantitative evaluation metrics — no accuracy rate, miss rate, false-alarm rate, ROC curve, or comparison table against baseline methods.
Equipment plausibility check (Section 1, Page 49): The Paxscan 2520 amorphous-silicon digital detector and GE CCO 2500-series X-ray source are consistent with equipment widely used in industrial NDT prior to and during 2014. No temporal inconsistencies were detected.

Evidence highlights

Direct textual contradiction between Equation (2) (median filter formula) and the subsequent sentence describing use of the "灰度平均值" (grayscale mean value) for replacement.
Absence of any numerical performance table or statistical comparison in Section 3 (experimental description) and Section 4 (conclusion).
Equipment specifications cited (127 μm pixel pitch, built-in 12-bit A/D) match standard specifications of the Paxscan 2520 platform.
DOI: 10.6040/j.issn.1672-3961.0.2014.013

Notes

This review is based solely on extracted text; original figures (Figures 1–5) and raw experimental data were not supplied, so image-reuse, splicing, and statistical distribution anomalies could not be assessed.
The median-vs.-mean description error is categorized as academic sloppiness rather than deliberate data fabrication, but it remains a violation of rigorous scholarly writing.
Follow-up actions suggested: obtain the original PDF for pixel-level inspection of Figures 2–5, and request raw experimental data and source code from the authors to verify the claimed accuracy.