Integrity Review Report: Environmental Sound Classification Based on Parallel Convolution and High-Low Frequency Attention

Academic fraud report · Geng Detector

Summary

This report assesses the paper "基于并联卷积和高低频注意力的环境声音分类" (DOI: 10.16208/j.issn1000-7024.2026.03.024), published in Computer Engineering and Design (2026, Vol. 47, No. 3), authored by Xu Jieke, Ge Dongyuan, Yao Xifan, and Li Yibo. The overall verdict is "Highly Suspicious" based on multiple textual and methodological anomalies, though pixel-level image analysis was not possible. The most serious concern (Finding 1) involves a comparison experiment in Table 4 citing reference [22] (Gao Y et al., on radar jamming pattern recognition via hyperspherical triplet coding) as achieving 96.03% accuracy on UrbanSound8K — a dataset of urban sounds such as dog barks and sirens — which is implausible and suggests fabrication or improper copying of comparison data. Additional issues include a poorly justified cross-domain citation for Self-gating (a breast cancer histopathology paper), an incorrect postal code (515000 belongs to Shantou, not Guangzhou where South China University of Technology is located), an irregular 15-month revision timeline, and absence of mean ± standard deviation reporting. These issues warrant formal editorial investigation. Confidence is moderate; limitations include no access to figures or raw data.

Verdict

Highly Suspicious — Multiple serious textual, citation, and methodological anomalies suggest possible data fabrication or scholarly negligence. Image-level duplication could not be assessed.

Key findings

Implausible cross-domain benchmark comparison (Table 4, Ref [22]): The paper compares its model against Gao Y. et al.'s "Jamming pattern open set recognition based on hyperspherical triplet coding" — a radar/electronic-warfare signal paper — claiming 96.03% accuracy on UrbanSound8K. UrbanSound8K contains 10 classes of urban sounds (e.g., dog barking, sirens, street music). It is highly implausible that a radar jamming paper was evaluated on this dataset, raising strong suspicion of fabricated comparison results or improper copying of tables from another paper.
Cross-domain methodological citation (Section 1.1.3, Ref [12]): Self-gating mechanism is anchored to Xie L. et al.'s "SHISRCNet: Super-resolution and classification network for low-resolution breast cancer histopathology image" — an unsuitable primary theoretical source for a gating formula in an audio classification model.
Incorrect postal code / geographic error: The second author Yao Xifan is listed with affiliation "South China University of Technology, Guangzhou, 515000". The postal code 515000 belongs to Shantou city, not Guangzhou (SCUT Wushan campus postal code is 510641). The error spans ~hundreds of kilometers.
Irregular publication timeline: Received 2024-11-23, revised 2026-02-27, published March 2026 — a 15-month revision cycle followed by publication within ~1 month of final revision.
Missing statistical rigor in deep learning experiments: All reported accuracies (e.g., 97.08%, 96.81%) are single point estimates without mean ± standard deviation across multiple runs, which is atypical for rigorous deep learning evaluation.
Image-based checks not performed: No figure pixel data was available, so image reuse and Photoshop-based manipulation could not be assessed.

Evidence highlights

DOI: 10.16208/j.issn1000-7024.2026.03.024
Reported comparison value under dispute: 96.03% (Table 4, attributed to Ref [22])
Disputed postal code: 515000 (Shantou, not Guangzhou)
Timeline: Received 2024-11-23, revised 2026-02-27, published March 2026 (Vol. 47, No. 3)
Ref [22] title: *Jamming pattern open set recognition based on hyperspherical triplet coding* (radar / electronic warfare domain)
Ref [12] title: *SHISRCNet: Super-resolution and classification network for low-resolution breast cancer histopathology image* (medical imaging domain)

Notes

All findings are derived from textual and structural inspection only; no figure-pixel, raw-data, or code verification was possible.
The cross-domain comparison in Finding 1 is the strongest indicator of potential misconduct and should be the priority of any formal investigation.
Postal-code and citation errors alone do not constitute fabrication, but in combination they point to weak scholarly care or possible ghostwriting.
A formal request to the authors for source code, training logs, and the source of the UrbanSound8K benchmark numbers for Ref [22] is recommended.