Verified Rigor RecordRigor Reviewed

Rigor-Reviewed Manuscript

Archimedes optimisation algorithm quantum dilated convolutional neural network for road extraction in remote sensing images · 10.1016/j.heliyon.2024.e26589

Name: Verified Rigor Record — Archimedes optimisation algorithm quantum dilated convolutional neural network for road extraction in remote sensing images · 10.1016/j.heliyon.2024.e26589
Item: Archimedes optimisation algorithm quantum dilated convolutional neural network for road extraction in remote sensing images · 10.1016/j.heliyon.2024.e26589
Rating: 67
Author: Αlpha¹

Paper source

Αlpha¹ Editorial

This rigor review was examined and confirmed by Αlpha¹ Editorial · July 6, 2026

Αlpha¹ Rigor Review

Rigor Reviewed

WARN

Warnings to address · Computational Study

Eight dimensions — click a segment for details

Critical

High-priority

Model-reported confidence: moderateReviewer agreement: lowEngine: 7.5.0Source: full text

Not verified in this report

References were not verified against Crossref/OpenAlex.
No reported statistical tests were found to recompute.

This Αlpha¹ Rigor Review uses AI Rigor Reviewers trained on a curated corpus of high-fidelity and retracted papers, with expert supervision and curation. It can still make mistakes; verify each finding against the source before relying on it.

The paper proposes a deep learning method (AOA-QDCNNRE) for road extraction and reports high performance on benchmark datasets. However, the study has significant reproducibility gaps: no software framework or code is provided, no training hyperparameters are reported, and the data availability statement has dead links. The quantum convolution claim is unsubstantiated. Statistical reporting is limited to point estimates without variability or inferential tests. The paper requires substantial improvements in reporting and evidence before it can be considered robust.

This is a post-publication audit of a computational study. The evaluation focused on the 6 applicable rigor dimensions (biological variables and ethical approvals are N/A). All reviewers converged on most dimensions, with minor divergences resolved by majority and checklist rule adherence. The statistics coverage note applies: no inferential tests were checked. The integrity check flagged a potential method/result mismatch regarding the quantum claim; claim audit identified two under-evidenced claims.

11 major claims checked against the paper's own evidence: 2 not fully backed by the presented evidence (unsupported or overstated).

3 integrity concerns flagged (0 high).

Rigor

Eight Dimensions1 pass · 4 warn · 1 fail

11 major claims1 unsupported1 overstated3 partial6 supported

Prior work on road extraction using deep learning is cited, including U-Net, attention mechanisms, and deformable convolutions. The limitations of conventional CNNs in capturing fine-grained details and contextual information are explicitly stated. The proposed AOA-QDCNNRE technique directly addresses these limitations by integrating quantum dilated convolutions. However, the paper does not explicitly discuss or address limitations of the cited prior work in detail.

Sub-criteria

prior work citedADEQUATE

premise rationaleADEQUATE

limitations addressedINADEQUATE

Per reviewer (3) — the votes behind the verdict

PASSReviewer 1· 85% conf

The paper adequately cites prior work and establishes a clear rationale for the proposed method, though it does not deeply analyze limitations of prior research.

Evidence

direct quote[Introduction, paragraphs 3-5]

“Conventional convolutional neural networks (CNNs) achieve remarkable results in road extraction tasks. The research gap in the existing works are often struggle with capturing fine-grained details and contextual information.”

direct quote[Introduction, paragraph 6]

“The AOA-QDCNNRE technique is mainly based on the QDCNN model, which integrates the concept of quantum computing (QC) with dilated convolutions for enhancing the network's capability to capture either local or global contextual data.”

paraphrase[Introduction, paragraph 5]

“The paper mentions that existing methods 'still could not retain boundary data and attain high-resolution road segmentation maps' but does not analyze why or how the proposed method specifically overcomes these limitations.”

PASSReviewer 2· 90% conf

The paper clearly establishes a research gap (conventional CNNs struggle with fine details and contextual information), cites relevant prior work, and provides a logical rationale for the proposed approach.

Evidence

direct quote[Introduction (paragraphs 4-7)]

“Shao et al. [] created a road extraction network that integrates an embedded attention system... Li et al. [] devise a method for extracting boundary-refined roads from RSI...”

direct quote[Introduction (paragraph 3)]

paraphrase[Introduction]

“The paper introduces AOA-QDCNNRE as a novel method to address the challenges of road extraction but does not map specific prior limitations to specific design choices.”

PASSReviewer 3· 90% conf

The paper cites prior work on road extraction using deep learning and establishes a rationale for the proposed method, though it does not explicitly discuss limitations of prior work in depth.

Evidence

direct quote[Introduction, paragraph 2 and 4-5]

absence[Introduction, paragraph 4, and Section 2]

The paper does not explicitly discuss how the proposed method addresses specific limitations of prior work beyond the general statement of improvement.

direct quote[Introduction, paragraph 6]

The study evaluates the AOA-QDCNNRE method on two public datasets (Massachusetts Roads and GF-2 Road). This constitutes a form of inclusion_exclusion (dataset selection) and independent_replication (two datasets). However, no explicit train/validation/test split percentages are reported, no mention of cross-validation or repeated runs, and no description of data preprocessing (e.g., tiling, normalization, augmentation). Power_analysis, in the sense of justifying the sample size (number of images or patches), is not addressed. Controls, in the form of baseline comparisons to multiple other models, are provided. Outlier_handling is not discussed. Independent_replication is partially covered by using two separate datasets.

Sub-criteria

controlsADEQUATE

power analysisNOT REPORTED

blinding levelsN/A

outlier handlingNOT REPORTED

randomization unitN/A

inclusion exclusionINADEQUATE

randomization methodN/A

replicate distinctionN/A

independent replicationINADEQUATE

Per reviewer (3) — the votes behind the verdict

WARNReviewer 1· 80% conf

The study uses standard benchmark datasets and compares against multiple baselines, but lacks justification for dataset choice and does not report data curation details or a formal power analysis.

Evidence

absence[Section 3, Results and discussion]

No power analysis or dataset selection justification.

absence[Section 3, Results and discussion]

Data curation or inclusion/exclusion criteria are not described.

direct quote[Tables 1 and 2]

“Comparative outcome of AOA-QDCNNRE approach with DL techniques on the Massachusetts Road dataset.”

WARNReviewer 2· 85% conf

The study uses two benchmark datasets to evaluate performance, which is appropriate for a computational paper, but lacks many design details expected for rigorous evaluation, such as explicit train/validation/test splits, cross-validation strategy, and description of data preprocessing.

Evidence

direct quote[Section 3 (Results and discussion), paragraph 1]

“the AOA-QDCNNRE method is executed on 2 datasets, namely, Massachusetts Road [] and GF-2 Road [] Dataset.”

absence[Section 2 (The proposed model) and Section 3 (Results and discussion)]

Data splits and preprocessing are not described.

paraphrase[Table 1 and Table 2]

“The AOA-QDCNNRE method is compared against CNN, U-Net, GL-Dense-U-Net, RDRCNN, and RDRCNN+post-process models.”

PASSReviewer 3· 85% conf

The design is a computational method with an evaluation on two benchmark datasets. Most design sub-criteria are either not applicable or adequately addressed for this study type.

Evidence

direct quote[Section 2, paragraphs 2.1 and 2.2]

“The QDCNN model presents a road extraction methodology that incorporates a quantum layer, convolutional layer, and dilation layer in a three-tier-based framework.”

direct quote[Section 3, first paragraph]

“the road extraction outcome of the AOA-QDCNNRE method is executed on 2 datasets, namely, Massachusetts Road and GF-2 Road Dataset.”

absence[Sections 2 and 3]

No explicit criteria for data inclusion/exclusion or outlier handling are reported.

No organisms, cells, or human subjects are used. The datasets are satellite imagery. Therefore, all biological variables sub-criteria are not applicable.

Sub-criteria

demographicsN/A

sex reportedN/A

sex justifiedN/A

age weight healthN/A

housing conditionsN/A

species strain sourceN/A

Per reviewer (3) — the votes behind the verdict

N/AReviewer 1· 100% conf

The study uses remote sensing images and does not involve biological subjects.

Evidence

direct quote[Ethics declarations]

“This article does not contain any studies with human participants or animals performed by any of the authors.”

N/AReviewer 2· 100% conf

The study uses remote sensing images of roads, not biological organisms, so biological variables are not applicable.

Evidence

direct quote[Title and Abstract]

“Archimedes optimisation algorithm quantum dilated convolutional neural network for road extraction in remote sensing images”

N/AReviewer 3· 95% conf

The study is purely computational using benchmark satellite image datasets, involving no biological subjects or human participants.

Evidence

direct quote[Abstract and Section 1]

“Roads are closely intertwined with human existence, and the process of extracting road networks has emerged as the most prominent task in remote sensing (RS).”

direct quote[Section 3, first paragraph]

“the road extraction outcome of the AOA-QDCNNRE method is executed on 2 datasets, namely, Massachusetts Road and GF-2 Road Dataset.”

The paper explicitly states no human or animal studies were performed. The datasets (Massachusetts Roads, GF-2) are publicly available. Therefore, ethics approval is not applicable.

Sub-criteria

iacuc statementN/A

informed consentN/A

irb ethics statementN/A

regulatory complianceN/A

Per reviewer (3) — the votes behind the verdict

N/AReviewer 1· 100% conf

No human or animal subjects, and no non-public subject-level data are used; the datasets are publicly available benchmark datasets.

Evidence

direct quote[Ethics declarations]

“This article does not contain any studies with human participants or animals performed by any of the authors.”

N/AReviewer 2· 100% conf

The study uses only public, pre-existing benchmark datasets; it does not involve new data collection from human or animal subjects, so ethics approvals are not applicable.

Evidence

direct quote[Ethics declarations]

“This article does not contain any studies with human participants or animals performed by any of the authors.”

direct quote[Data availability statement]

“Databases are available in the following open access datasets.”

PASSReviewer 3· 95% conf

The study uses only public, fully de-identified benchmark image datasets, so ethics approval and consent are not applicable. The paper includes a statement that no human or animal studies were conducted.

Evidence

direct quote[Ethics declarations section]

“This article does not contain any studies with human participants or animals performed by any of the authors.”

direct quote[Data availability statement]

“Datasets are available in the following open access datasets. https://www.kaggle.com/datasets/balraj98/massachusetts-buildings-dataset ...”

For a computational study, only software_tools_identified is applicable. The paper describes the AOA-QDCNNRE method but does not name the deep learning framework (e.g., TensorFlow, PyTorch), version, or any custom code repository. No software versions or URLs are given. All other sub-criteria (antibodies, cell lines, etc.) are not applicable.

Sub-criteria

mycoplasma testingN/A

reagents identifiedN/A

organisms identifiedN/A

antibodies identifiedN/A

cell line authenticationN/A

software tools identifiedNOT REPORTED

Per reviewer (3) — the votes behind the verdict

FAILReviewer 1· 90% conf

The paper does not identify any software tools, libraries, or frameworks used, and no code is provided, making reproducibility impossible.

Evidence

absence[Sections 2 and 3]

No software tools or frameworks are identified.

absence[Data availability statement]

No code repository is provided.

WARNReviewer 2· 85% conf

The software tools are partially identified, but key model hyperparameters and implementation details are missing. The datasets are identified with URLs.

Evidence

direct quote[Data availability statement]

“https://www.kaggle.com/datasets/balraj98/massachusetts-buildings-dataset https://www.cs.toronto.edu/∼vmnih/data/mass_roads/train/sat/index.html https://paperswithcode.com/dataset/gid”

absence[Section 2 and Section 3]

Software tools (deep learning framework) are not identified by name and version.

absence[Section 2 and Section 3]

Training hyperparameters (learning rate, batch size, epochs) are not reported.

PASSReviewer 3· 85% conf

The only applicable resources are the software/tools used for the computational method, which are partially identified. No biological reagents, antibodies, cell lines, or animals are used.

Evidence

absence[Sections 2.1 and 2.2]

The paper describes the QDCNN model and AOA algorithm but does not name the software framework (e.g., Python, TensorFlow, PyTorch) or provide version numbers.

absence[Data availability statement and entire paper]

No code repository or link is provided.

direct quote[Ethics declarations and Methods]

“This article does not contain any studies with human participants or animals performed by any of the authors.”

Performance metrics are reported for the proposed method and baselines on two datasets. The tests are not named (e.g., no statistical test is used to compare models). Assumptions are not verified. Exact p-values are not reported; effect sizes and confidence intervals are absent. The statistical software is not named. For data presentation, total accuracy, precision, recall, F1, and IoU are given as single point estimates without variability (no SD, SEM, or CI), though this is common in computational benchmarking. The metrics are presented in tables and bar charts. Mathematical plausibility is not applicable because the reported metrics are proportions, but the paper does not provide per-class counts or raw confusion matrices to verify the arithmetic.

Sub-criteria

tests namedNOT REPORTED

exact p valuesNOT REPORTED

effect sizes ciNOT REPORTED

data presentationINADEQUATE

software identifiedNOT REPORTED

assumptions verifiedN/A

mathematical plausibilityN/A

Per reviewer (3) — the votes behind the verdict

N/AReviewer 1· 95% conf

The paper does not perform any inferential statistical tests; it only reports descriptive metrics (accuracy, precision, recall, F1, IoU).

Evidence

absence[Section 3]

No statistical tests are named or used.

direct quote[Tables 1, 2, 3]

“A c c u y, P r e c n, R e c a l, F 1 S c o r e, IoU”

WARNReviewer 2· 90% conf

The paper reports performance metrics (accuracy, precision, recall, F1, IoU) for model comparisons, but lacks any inferential statistics (no confidence intervals, hypothesis tests, or standard deviations). The mathematical plausibility check is not applicable as only aggregate metrics are reported.

Evidence

absence[Section 3 (Results and discussion)]

No statistical tests are used to compare models.

direct quote[Table 1 and Table 2]

“AOA-QDCNNRE: 99.72% accuracy, 99.30% precision, 99.63% recall, 99.13% F1 score, 99.33% IoU”

absence[Section 2 and Section 3]

No software environment is identified for the experiments.

WARNReviewer 3· 80% conf

The paper reports accuracy, precision, recall, F1, and IoU metrics for model comparisons, but does not name statistical tests, verify assumptions, report exact p-values, or provide effect sizes/confidence intervals. No inferential statistics are performed.

Evidence

direct quote[Section 3, Tables 1-3 and accompanying text]

“Table 1 Comparative outcome of AOA-QDCNNRE approach with DL techniques on the Massachusetts Road dataset.”

absence[Section 2 and 3]

Software/framework used is not identified.

direct quote[Section 3, figures 6,7,10,11 and text]

“Fig. 6 Accuracy curve of AOA-QDCNNRE method on Massachusetts Road dataset.”

The paper provides a data availability statement naming three URLs for the two benchmark datasets (Massachusetts Roads and GID). These are adequate for the input data. However, the AOA-QDCNNRE method's implementation (custom code) is not shared; no repository, DOI, or supplementary material is provided. The paper also does not report any accession numbers for generated data or predictions. For a computational paper, code sharing is a key expectation.

Sub-criteria

code sharingNOT REPORTED

accession numbersNOT REPORTED

repository depositNOT REPORTED

data availability statementADEQUATE

Per reviewer (3) — the votes behind the verdict

WARNReviewer 1· 85% conf

The data availability statement provides URLs to the benchmark datasets, but no code is shared, and no custom data or code repository is provided.

Evidence

direct quote[Data availability statement]

“Datasets are available in the following open access datasets. https://www.kaggle.com/datasets/balraj98/massachusetts-buildings-dataset ... https://paperswithcode.com/dataset/gid”

absence[Data availability statement and entire paper]

No code is shared.

WARNReviewer 2· 90% conf

A data availability statement is present and names the public datasets, which is adequate for those external resources. However, no code is provided, and no script or custom implementation is shared. The output data (predicted segmentation maps) are not deposited.

Evidence

direct quote[Data availability statement]

“Databases are available in the following open access datasets. https://www.kaggle.com/datasets/balraj98/massachusetts-buildings-dataset ... https://paperswithcode.com/dataset/gid”

absence[Data availability statement, Section 5 (Conclusion)]

No code repository is provided.

absence[Data availability statement]

No accession numbers for generated data or outputs.

PASSReviewer 3· 90% conf

A data availability statement is present with concrete URLs for two of the three datasets used. No custom code repository is provided. For a computational study, code sharing is expected but the paper provides detailed formulas for the method.

Evidence

direct quote[Data availability statement]

“Datasets are available in the following open access datasets. https://www.kaggle.com/datasets/balraj98/massachusetts-buildings-dataset https://www.cs.toronto.edu/∼vmnih/data/mass_roads/train/sat/index.html https://paperswithcode.com/dataset/gid”

absence[Entire paper]

No code repository or link to custom implementation is provided.

direct quote[Section 2, equations 1-22]

“O_i = l b_i + r a n d × ( u b_i - l b_i)”

Methods are provided mathematically but are incomplete: no training hyperparameters (learning rate, batch size, number of epochs, optimizer), no hardware details, no software framework. Trial registration and reporting guideline are not applicable. Limitations are discussed in one sentence: 'The limitation of the AOA-QDCNNRE approach is only tested in simulation experiments.' All pre-specified outcomes (accuracy, precision, recall, F1, IoU) are reported for both datasets. Conclusions are proportional: the proposed method 'surpasses recent algorithms' on the tests conducted. Funding and conflict of interest are clearly stated.

Sub-criteria

funding coiADEQUATE

trial registrationN/A

reporting guidelineN/A

methods completenessINADEQUATE

all outcomes reportedADEQUATE

limitations discussedINADEQUATE

conclusions proportionalADEQUATE

Per reviewer (3) — the votes behind the verdict

PASSReviewer 1· 80% conf

The paper reports most applicable criteria adequately, including limitations, proportional conclusions, and funding, but methods lack sufficient detail for full replication.

Evidence

direct quote[Section 2, The proposed model]

“The proposed model encompasses two major processes such as QDCNN-based road extraction and AOA-based hyperparameter tuning.”

direct quote[Section 4, Conclusion]

“The limitation of the AOA-QDCNNRE approach is only tested in simulation experiments.”

direct quote[Funding and Declaration of competing interest]

“The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.”

WARNReviewer 2· 90% conf

The methods section describes the model architecture and optimization algorithm, but lacks critical details for independent replication (e.g., training hyperparameters, hardware, environment). Limitations are only briefly mentioned. Conclusions are proportional. Funding and COI are reported.

Evidence

paraphrase[Section 2 (The proposed model)]

“The QDCNN architecture and AOA algorithm are described with equations, but hyperparameters like learning rate, batch size, and epochs are not specified.”

direct quote[Section 5 (Conclusion)]

“The limitation of the AOA-QDCNNRE approach is only tested in simulation experiments.”

direct quote[Funding, Declaration of competing interest]

“This research has been funded by the Deanship for Research & Innovation, Ministry of Education in Saudi Arabia... The authors declare that they have no known competing financial interests...”

WARNReviewer 3· 85% conf

The methods are described in detail, limitations discussed, conclusions are proportional to the results, and funding/COI statements are provided. However, no trial registration or reporting guideline is applicable, and the paper does not explicitly report all outcomes (focuses on positive results).

Evidence

direct quote[Section 4, Conclusion]

“The limitation of the AOA-QDCNNRE approach is only tested in simulation experiments.”

direct quote[Funding and Declaration of competing interest sections]

“This research has been funded by the Deanship for Research & Innovation, Ministry of Education in Saudi Arabia, for funding this research work through project number: IFP22UQU4281768DSR120.”

direct quote[Section 3, Tables 1-2]

“AOA-QDCNNRE technique exhibited results with a maximum a c c u y of 99.72%, p r e c n of 99.30%, r e c a l of 99.63%, F1 s c o r e of 99.13%, and IoU of 99.33%.”

23 copyedit issues flagged (12 major): mostly typo, grammar, clarity.

Reproducibility

4 data/code links checked; 2 live, 2 dead.

Αlpha¹ Rigor Action ItemsFinal pass before journal submission

The published paper has substantial reporting gaps that an informed reader should weigh: the lack of code, software framework details, and training hyperparameters make independent replication impossible. The unsupported quantum convolution claim and the unusually high point metrics without variability warrant caution. The dead data links further undermine verification. An erratum providing code and training details, and clarifying the quantum implementation, would be needed before the paper can be considered robust. The paper is not suitable as a standalone evidence source for the claims made.

1.
HIGHreporting
Section 2 (The proposed model): report the deep learning framework and version (e.g., TensorFlow 2.x, PyTorch 1.x) and the hardware used (GPU model, memory) to enable reproducibility.
Without the software framework and hardware specifications, independent researchers cannot replicate the experiments.
2.
HIGHdata code
Data availability statement: provide a link to a public code repository (e.g., GitHub, Zenodo) with a DOI for the custom AOA-QDCNNRE implementation.
Code sharing is the expected standard for computational papers; its absence prevents verification and reuse of the method.
3.
HIGHrigor
Section 2 (The proposed model): clarify the quantum convolution implementation — specify whether it is simulated or run on actual quantum hardware, provide a reference for the quantum circuit, and include an ablation study isolating the quantum component's contribution.
The paper claims quantum enhancement but provides no evidence; without a clear description, the claim is unsupported and potentially misleading.
4.
HIGHstatistics
Section 3 (Results): report variability measures (standard deviation or confidence intervals) across multiple training runs (e.g., 5 runs with different random seeds) for all performance metrics.
Point estimates without variability cannot be assessed for stability or statistical significance, undermining confidence in the reported scores.
5.
HIGHreporting
Section 2 (The proposed model): report all training hyperparameters (optimizer, learning rate, batch size, number of epochs, learning rate schedule, early stopping criteria).
These hyperparameters are essential for replication and for assessing the likelihood of overfitting.
6.
HIGHother
Section 3 (Experimental setup): justify the selection of datasets (e.g., sample size, resolution, diversity) and describe data preprocessing steps (tiling size, normalization, augmentation).
Dataset selection and preprocessing directly affect model performance and generalizability; their omission makes the evaluation framework incomplete.
7.
HIGHother
Section 3 (Experimental setup): explicitly state the train/validation/test split percentages or describe the cross-validation strategy used.
Without the split information, it is impossible to evaluate whether the reported metrics are from an independent test set or potentially overfit.
8.
HIGHdata code
Data availability statement: fix the dead links (the reproducibility check found two of four URLs dead) or provide alternative working URLs.
Broken data links prevent readers from accessing the benchmark datasets, undermining the data availability statement.
9.
HIGHreporting
Section 4 (Conclusion): expand the limitations section beyond 'only tested in simulation experiments' to address potential overfitting, sensitivity to dataset characteristics, failure modes, and computational cost.
A brief limitations statement is inadequate for readers to interpret the scope and generalizability of the results.
10.
HIGHreporting
Section 3 (Results): include an ablation study comparing AOA-QDCNNRE with and without the AOA tuning, and with standard dilated convolutions without quantum components, to support the claimed benefits.
The claim audit found that two headline claims (AOA hyperparameter tuning improves outcomes; quantum integration enhances contextual capture) are unsupported without ablation experiments.
11.
MEDIUMstatistics
Section 3 (Results): use statistical tests (e.g., McNemar's test for paired comparisons or bootstrapping) to assess whether performance differences between AOA-QDCNNRE and baselines are statistically significant.
Without statistical tests, apparent improvements may be due to random chance rather than genuine method superiority.
12.
MEDIUMreporting
Abstract and Introduction: correct subject-verb agreement errors and clarify phrasing (e.g., 'The research gap in the existing works are often struggle with...' to 'The research gap is that existing works often struggle with...').
Grammatical errors undermine readability and professional presentation.
13.
MEDIUMcopyedit
Tables 1, 2, 3: replace corrupted column headers (e.g., 'A c c u y, P r e c n, R e c a l') with standard terms 'Accuracy', 'Precision', 'Recall'.
The current headers are difficult to read and suggest a formatting error that should be corrected for clarity.
14.
LOWother
Remove or integrate the references to unrelated engineering systems (CCHP, SOFC, PEMFC, HRES) that appear to be padding.
These references do not contribute to the road extraction method and may reduce the perceived focus and relevance of the paper.
15.
LOWdata code
Deposit the trained model weights or predicted segmentation maps in a public repository (e.g., Zenodo, Hugging Face) to allow verification of the reported metrics.
Even with code, model weights or outputs enable direct comparison and verification of results.

Full text (35,218 chars)Analyzed: Jul 6, 2026

Αlpha¹ assesses methodological rigor, not the importance of the findings. See how we calibrate →