Integrity Assessment Report: Welding Defect Detection Algorithm Based on Feature Extraction and Extremum Search (Liang et al., 2014)

Academic fraud report · Geng Detector

Summary

This report evaluates the 2014 paper by Liang Wei, Tao Liang, Zhang Guangxian, and Li Zhenhua, published in the Journal of Shandong University (Engineering Science), DOI 10.6040/j.issn.1672-3961.0.2014.013. Overall verdict: highly suspicious on grounds of paper quality and internal logic. Two substantive issues were identified. First, the frame-averaging formula in Section 2.1 (Eq. 1) contains a fundamental dimensional and logical error: the left-hand side D(i,j) is a per-pixel gray value, while the right-hand side sums over all pixel coordinates (i=1..h, j=1..w) and frames, yielding a scalar constant, not a pixel value. This suggests formula fabrication, careless transcription, or copy-paste from an unrelated source. Second, the conclusion claims 'high detection accuracy' but the paper provides no quantitative metrics such as precision, recall, or error rates; results are described only with qualitative phrases. Image-reuse and statistical checks were not possible because no source images or raw data tables were supplied. Equipment timeline is consistent. Confidence is limited to textual and logical analysis only; pixel-level and statistical findings remain undetermined.

Verdict

Highly suspicious (orange). Based solely on textual and logical analysis — no pixel-level image forensics were performed.

Key findings

Critical mathematical error in core formula (Section 2.1, Eq. 1). The frame-averaging denoising formula is dimensionally and logically inconsistent, suggesting formula fabrication, careless copying, or content assembled without genuine understanding.
Unsupported quantitative claim of 'high detection accuracy' (Section 3.2 and Section 4 conclusion). No accuracy, precision, recall, or error metrics are reported anywhere in the paper; experimental results are described only with subjective phrases.
Equipment timeline consistent with 2014 publication date. Paxscan 2520 amorphous-silicon digital detector and GE CCO 2505 portable X-ray unit were standard industrial equipment available before 2014.
Image-reuse and statistical checks not possible — source images and raw data tables were not provided.

Evidence highlights

Equation 1 as written: D(i,j) = (1/m) · Σ_{i=1}^{h} Σ_{j=1}^{w} Σ_{t=1}^{m} F_t(i,j). The double sum over all image coordinates collapses the right-hand side to a scalar (global mean of the entire image), which cannot equal a single per-pixel gray value D(i,j). The correct expression should be D(i,j) = (1/m) · Σ_{t=1}^{m} F_t(i,j).
Quantitative metrics absent: no tables report accuracy, recall, or error figures; descriptive wording includes only "趋于平稳" (Figure 2) and "使焊缝更加清晰" (Figure 4).
Figure caption list (Figure 1 system composition; Figure 2 noise-residual curve; Figure 3 seam segmentation; Figure 4 histogram equalization; Figure 5 waveform analysis) is internally consistent with a typical digital image processing paper; no textual contradictions noted.

Notes

Scope limit: this assessment is based on the PDF text only. Image-manipulation (PS traces, reuse), Western-blot forensics, and statistical distribution checks could not be performed.
Reference DOI preserved exactly: 10.6040/j.issn.1672-3961.0.2014.013.
Given that the paper is 12 years old and the suspected issues concern rigor rather than outright data falsification, a formal journal complaint is not recommended; the case is more useful as a teaching example for academic-integrity training.
All findings are uncertain until the authors provide source code and the original image test set for independent verification.