Hybrid Machine Learning for Automated Road Safety Inspection of Auckland Harbour Bridge

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors In this paper, a hybrid machine learning method is used to solve the safety anomaly detection of the concrete section of the MCB on the Auckland Harbour Bridge. The research content of this manuscript is the application research of image recognition, which is interesting. The methods proposed in this paper are not clearly stated, and the contributions made are not well summarized. The overall completion of the manuscript is good, but there are still some contents that need to be further improved.

1. It is suggested that the innovation points of this paper should be summarized in the "Introduction" part and given directly.   Thus, readers can better understand the paper.

2.This paper proposes to obtain the safety detection results of MCB on bridge automatically by deep learning method, which has certain new ideas.   However, the proposed method is not described in detail in this paper, and the experimental results of several popular algorithms are compared.   Whether the author thinks that the main contribution of this paper is to establish a privacy-preserving automated monitoring system, that is transferrable from data collected on Auckland Harbour Bridge to similar contexts involving traffic flow regulation and safety monitoring applications? These need to be elaborated clearly, especially the innovation of the paper and the main contribution of this paper.

3. Some of the pictures in the article are fuzzy and need to be replaced. Such as Figure 8, 13.

4. There are many problems in the layout format of the paper, such as too much blank space in the paper(Pages at 7, 9 11, 17), too large pictures, and the tables are not properly formatted.

5. The discussion in the experimental part of the paper should be more logical, and it is suggested to compare and analyze with the latest methods to highlight the effectiveness of the proposed method.

6.The references of the paper are too old and need to be updated, and there is only one reference in 2024.

7.The whole article is too miscellaneous, it is suggested to delete unnecessary content and further clarify the logical relationship.

Comments on the Quality of English Language

Perfect

Author Response

Please note that I have used individual colors in the manuscript text to highlight some of the important or specific changes related to your feedback. The responses related to Reviewer 1 are highlighted in red, the responses related to Reviewer 2 are highlighted in green, and the responses related to Reviewer 3 are highlighted in purple. Some of the text has mixed colors to indicate that it relates to feedback from multiple reviewers.

1. It is suggested that the innovation points of this paper should be summarized in the "Introduction" part and given directly.   Thus, readers can better understand the paper.

Response: Thank you for your valuable feedback, which has helped us improve the clarity and depth. To address this feedback, we have summarized the innovation points of our paper in the "Introduction" section (Line 67 to 73 and Line 95 to 112) to enhance clarity and ensure that readers can quickly grasp the key contributions of our research. The revised introduction now includes the following summary of innovation points:

The study developed a hybrid machine learning system for real-time, privacy-preserving anomaly detection in road safety inspections. The research's contributions are listed as follows:

• Produces a traffic safety analysis artefact scalable to scenarios in over 20+ countries and hundreds of similar traffic scenes [10] that employ movable concrete barriers.
• Introduces a semi-automated synthetic data generation method using a novel background cloning technique. The novel approach addresses data sparsity and enhances model training, with repeatability value for other computer vision case studies facing dataset balancing issues.
• Refines classification methods to balance false positives and negatives, improving detection accuracy from 82.6% (reported in earlier peer-reviewed research [11]) to 93.6%. Achieving this 11% increase in accuracy within complex traffic scenes characterized by chaotic backgrounds and lighting conditions underscores the artefact's viability for real-world applications.
• Successfully navigates hazardous traffic scenes for data collection by adhering to industry safety protocols. This repeatable approach provides a comprehensive blueprint for managing similar scenarios, ensuring stakeholder satisfaction and achieving sufficient data.

 

2.This paper proposes to obtain the safety detection results of MCB on bridge automatically by deep learning method, which has certain new ideas.   However, the proposed method is not described in detail in this paper, and the experimental results of several popular algorithms are compared.   Whether the author thinks that the main contribution of this paper is to establish a privacy-preserving automated monitoring system, that is transferrable from data collected on Auckland Harbour Bridge to similar contexts involving traffic flow regulation and safety monitoring applications? These need to be elaborated clearly, especially the innovation of the paper and the main contribution of this paper.

Response: Thank you for your valuable feedback, which has helped us improve the clarity and depth of our manuscript. We have made the following changes to address the reviewer's comments:

1. Introduction Section:
• Added a summary of the main contributions and innovations to provide readers with a clear understanding of the paper’s key advancements (Line 67 to 73 and Line 95 to 112).
2. Materials and Methods Section:
• Provided a detailed description of the proposed method, including data collection, synthetic data generation, model training, and evaluation with step numbers for better clarity (Sections are reshuffled and edited).
3. Evaluation Section:
• Included a detailed comparison of the experimental results with previous baseline accuracy and other popular algorithms to highlight the improvements and effectiveness of the proposed method (table 8, line 557 to 581).

By making these changes, we have elaborated on the innovation and main contributions of our paper, ensuring clarity and comprehensiveness.

 

3. Some of the pictures in the article are fuzzy and need to be replaced. Such as Figure 8, 13.

Response:

We appreciate the reviewer's feedback regarding the clarity of the images. We have replaced and enhanced the quality of all relevant images in the manuscript, including Figures 8 and 13. These figures have been updated to ensure they are clear and of high resolution.

4. There are many problems in the layout format of the paper, such as too much blank space in the paper(Pages at 7, 9 11, 17), too large pictures, and the tables are not properly formatted.

Response: We appreciate the reviewer's feedback regarding the layout and formatting issues. We have made the following changes to address these concerns:

1. Reduction of Blank Spaces:
• Adjusted the layout to minimize blank spaces on pages 7, 9, 11, and 17.
2. Resizing Pictures:
• Resized all images to ensure they fit appropriately within the page margins and do not appear too large.
3. Table Formatting:
• Reformatted all tables to ensure they are properly aligned and fit well within the page layout without excessive blank space.

5. The discussion in the experimental part of the paper should be more logical, and it is suggested to compare and analyze with the latest methods to highlight the effectiveness of the proposed method.

Response: We appreciate the reviewer's suggestion to improve the logical flow and comparative analysis in the experimental section. To address this feedback, we have made the following improvements:

1. Enhanced Logical Flow:
• Reorganized the discussion in the experimental section to ensure a more logical and coherent flow of information. This includes a step-by-step explanation of the methodology, results, and their implications (added additional tables and figures).
2. Comparison with Latest Methods:
• Added a comparison table that includes the experimental results of the proposed method alongside several popular algorithms. This table highlights key performance metrics such as accuracy, precision, recall, F1-score, and ROC-AUC, providing a clear comparison.
• Included a ROC graph that visually compares the performance of the proposed method with the latest methods. This graph emphasizes the effectiveness of our approach.

These enhancements ensure that the experimental discussion is both logical and comprehensive, clearly demonstrating the superiority and effectiveness of our proposed method in comparison to other state-of-the-art techniques.

6.The references of the paper are too old and need to be updated, and there is only one reference in 2024.

Response: We appreciate the reviewer's feedback regarding the recency of the references, but we respectfully disagree on this feedback. Based on a detailed analysis, the majority of our references are indeed recent, with a significant number from the years 2023 (15 references) and 2024 (13 references). Here is a breakdown of the reference distribution:

• 1990s: 2 references (1991, 1994)
• 2000s: 1 reference (2009)
• 2010s: 2 references (2016, 2019)
• 2020: 9 references
• 2021: 8 references
• 2022: 6 references
• 2023: 15 references
• 2024: 13 references

Out of a total  about 74% of our references are 2020 onwards. This demonstrates that the bulk of our citations are indeed recent, ensuring that our paper is grounded in the latest research and developments in the field.

7.The whole article is too miscellaneous, it is suggested to delete unnecessary content and further clarify the logical relationship.

Response: We appreciate the reviewer's feedback regarding the structure and clarity of the article. To address this, we have made several revisions to improve the coherence and focus of the manuscript:

1. Content Streamlining:
• Removed Unnecessary Content: We carefully reviewed the manuscript and deleted or realigned sections and details that were redundant or did not directly contribute to the main objectives of the study.
2. Improved Logical Flow:
• Reorganized Sections: We restructured the paper to ensure a clear and logical progression of ideas. Each section now builds on the previous one, providing a coherent narrative from the introduction through the methodology to the results and discussion.
• Clearer Transitions: Enhanced the transitions between sections and subsections to guide the reader through the document smoothly.
3. Focused Discussions:
• Enhanced Discussion Section: Clarified and focused the discussion on key findings and their implications, ensuring that each point is directly relevant to the study’s objectives.
• Comparison with Existing Methods: Added detailed comparisons with existing methods to highlight the effectiveness and innovation of the proposed method, making the discussion more insightful and relevant.
4. Summarized Innovations:
• Introduction Section: Summarized the innovation points in the introduction to provide a clear overview of the paper’s contributions at the outset.

Comments on the Quality of English Language

Perfect

Response: Thank you

Reviewer 2 Report

Comments and Suggestions for Authors

The paper presents a machine-learning solution to enhance safety inspections of the movable concrete barrier (MCB) on the Auckland Harbour Bridge. The system uses video data to identify safety hazards, improving the initial classification accuracy from 82.6% to 93.6%. The authors aim to apply this method to other infrastructure inspections in the future. The work is interesting, and has valuable application; however, I have the following concerns, comments, and suggestions on the current version:

1. The abstract section needs improvement in terms of elaborating a little more on the techniques used in the manuscript. It should provide a comprehensive, holistic overview of the research paper.
2. The introduction section should also contain a brief overview of the work’s context in the literature, presenting some previous works in relation to the current work.
3. The contributions of the work need to be clearly stated.
4. The ‘Literature Review’ section is somewhat disorganized. It mixes literature studies with methodology and needs to be reorganized and refined.
5. In the ‘Materials and Methods’ section, I suggest adding a comprehensive pipeline that depicts the entire study steps and methods.
6. The ‘Materials and Methods’ section needs to be reorganized into subsections outlining each step of your methodology. Please maintain order according to the various steps of your pipeline. Additionally, a more in-depth theoretical description of the used methods should be added to the section.  
7. In Table 2, Pin_Out images exist only for Validation data, but not Training Data. Is this a mistake, or did you intentionally not train your model on Pin_Out images? An explanation might be required on this matter.
8. Some graphs in the figures need to be enhanced in terms of quality and labelling.
9. Failure cases and limitations of the research should also be discussed.

 

Overall, the manuscript is interesting, however, it still needs to be revised first. 

Comments on the Quality of English Language

 Minor editing of English language required

Author Response

Please note that I have used individual colors in the manuscript text to highlight some of the important or specific changes related to your feedback. The responses related to Reviewer 1 are highlighted in red, the responses related to Reviewer 2 are highlighted in green, and the responses related to Reviewer 3 are highlighted in purple.

1. The abstract section needs improvement in terms of elaborating a little more on the techniques used in the manuscript. It should provide a comprehensive, holistic overview of the research paper.

Response:

Thank you for your valuable feedback. To address this, we have added the following sentence to the abstract to elaborate on the techniques used (Line 23, 24):

 

"The study leverages the ResNet-50 based STENet architecture, integrated with 3D convolutions, to capture and analyze spatiotemporal data, facilitating real-time surveillance of the AHB."

1. The introduction section should also contain a brief overview of the work’s context in the literature, presenting some previous works in relation to the current work.

Response:

Thank you for your valuable feedback. To address this, we have revised the introduction section to include a brief overview of the context of the work in the literature and relevant previous works (Line 67 to 73). The updated introduction now includes the following summary:

Introduction (Revised)

Prior works in automated infrastructure inspection often fall short in dynamic and complex environments like the Auckland Harbour Bridge (AHB) [4-6]. This research aims to advance the field by incorporating a novel hybrid deep learning and spatiotemporal data analysis, allowing for more accurate and reliable detection of safety anomalies in complex environments such as the AHB. This study employs spatiotemporal analysis, deployable AI algorithms, and semi-automated synthetic data generation to enhance traffic barrier monitoring, transforming research into practical, real-time anomaly detection solutions.

1. The contributions of the work need to be clearly stated.

Response:

Thank you for your valuable feedback. To address this, we have clearly stated the contributions of our work in the introduction section. The revised introduction now includes the following summary of the contributions:

 

Introduction (Revised)

The study developed a hybrid machine learning system for real-time, privacy-preserving anomaly detection in road safety inspections (Line 95 to 112). The research's contributions are listed as follows:

The study developed a hybrid machine learning system for real-time, privacy-preserving anomaly detection in road safety inspections. The research's contributions are listed as follows:

 

• Produces a traffic safety analysis artefact scalable to scenarios in over 20+ countries and hundreds of similar traffic scenes [10] that employ movable concrete barriers.
• Introduces a semi-automated synthetic data generation method using a novel background cloning technique. The novel approach addresses data sparsity and enhances model training, with repeatability value for other computer vision case studies facing dataset balancing issues.
• Refines classification methods to balance false positives and negatives, improving detection accuracy from 82.6% (reported in earlier peer-reviewed research [11]) to 93.6%. Achieving this 11% increase in accuracy within complex traffic scenes characterized by chaotic backgrounds and lighting conditions underscores the artefact's viability for real-world applications.
• Successfully navigates hazardous traffic scenes for data collection by adhering to industry safety protocols. This repeatable approach provides a comprehensive blueprint for managing similar scenarios, ensuring stakeholder satisfaction and achieving sufficient data.

 

1. The ‘Literature Review’ section is somewhat disorganized. It mixes literature studies with methodology and needs to be reorganized and refined.

Response:

Thank you for your insightful feedback. To address this, we have reorganized and refined the ‘Literature Review’ section to clearly separate the literature studies from the methodology. The revised section now presents a more structured and coherent review of relevant literature, followed by a distinct methodology section with step number for clarity.

1. In the ‘Materials and Methods’ section, I suggest adding a comprehensive pipeline that depicts the entire study steps and methods.

Response:

Thank you for your valuable suggestion. To address this, we have restructured the section and also included step numbers to enhance clarity and guide readers through the process systematically.

1. The ‘Materials and Methods’ section needs to be reorganized into subsections outlining each step of your methodology. Please maintain order according to the various steps of your pipeline. Additionally, a more in-depth theoretical description of the used methods should be added to the section.  

Response:

Thank you for your valuable suggestion. To address this, we have restructured the section and also included step numbers to enhance clarity and guide readers through the process systematically.

1. In Table 2, Pin_Out images exist only for Validation data, but not Training Data. Is this a mistake, or did you intentionally not train your model on Pin_Out images? An explanation might be required on this matter.

Response:

Thank you for pointing out this issue. This was indeed a mixup with an unused table. We have corrected Table 3 to accurately reflect the distribution of Pin_Out images in both the training and validation datasets (Line 421 to 422).

1. Some graphs in the figures need to be enhanced in terms of quality and labelling.

Response:

Thank you for your valuable feedback. To address this, we have enhanced the quality and labelling of all graphs in the figures throughout the manuscript especially figure 8 and 13.

1. Failure cases and limitations of the research should also be discussed.

 Response:

Thank you for your feedback. We have updated the relevant sections and also would like to point out where already discussed the failure cases and limitations in multiple sections of the manuscript. Here are the key points:

Background Noise:

Detection accuracy is affected by background noise, making it erratic despite overall improvements.

Reference: Discussion on the need for a more diverse training dataset and how detection accuracy varies under different conditions.

Data from Different Angles:

The necessity for data from various angles is emphasized to further improve the model.

Reference: Mentioned in the context of enhancing robustness and accuracy by expanding data collection protocols.

Detailed Evaluation:

The evaluation of the STENet model includes configurations and performance metrics, highlighting the efficiency and challenges encountered.

Reference: Detailed in the performance evaluation section, comparing SGDM and Adam optimizers, and showing RMSE and validation losses (Table 8, 9, and 10).

Robust Performance with Limitations:

While the system performs well in ideal conditions, accuracy varies with different fields of view and lighting conditions.

Reference: Discussed the limitations in terms of dataset diversity and manual frame creation challenges.

 

 

Overall, the manuscript is interesting, however, it still needs to be revised first. 

 

Comments on the Quality of English Language

 Minor editing of English language required.

Response: Thank you, another proof reding is done to improve consistency.

Reviewer 3 Report

Comments and Suggestions for Authors

 Summary:

This paper introduces a hybrid deep machine learning method that combines kernel operation with customer transfer learning strategies. 

They use a multi-stage preprocessing to extract spatio-temporal region of interest with a rolling window to identify video frames containing diagnostic information.

The experiments show the improvements on a practical application.

Strengths:

+ the paper is overall technical sound

+ nice visualisations presented in the experimental section

+ some interesting comparisons and analysis presented in the paper

Weaknesses:

- One of the major issues with this submission is its novelty. The paper is more like exploring a combination of existing machine learning methods for the automation of road safety inspection.   - The paper reviewed some related works, e.g., advances in road defect and anomaly detection; however, the comparison regarding to existing methods are not analysed and discussed in the related work section.   - The method section should detail the algorithm proposed, in a clear way. The current paper, presents some equations, but the equations suffer from several issues, e.g., typos and issues with index, detailing below.   - The proposed method is not compared to existing methods, or the variant of the proposed model. There should be some variants study and ablation study to show the benefits of the model.   - The discussion section should be presented in a better way rather than listed in bullet points. The discussion section is also not clear and thorough, especially the discussions on limitations and future work.   - Table captions also not clear enough. Table 9 caption does not clearly show the information presented the paper with a proper explanation of the each column, etc.   - The evaluations are also quite limited. The authors claim that the proposed method addresses the issues of detection, but the detections regarding to multiple factors / defects are not clearly presented.   

- Some figures have issues with layouts, it would be better to rearrange them to make them look nicer.

- Eq. (2) seems to have issues with index, and maths symbols. The authors should use standard maths expressions, e.g., subscript or superscript etc. Issues also inside Eq. (4). The authors are encouraged to update all the maths equations to make them correct.

- It is suggested to have a notation section detailing the maths symbols used in the paper, e.g., regular fonts for numbers, lowercase bold face for vectors, and uppercase bold face for matrices etc.

- Tables could be rearranged nicely. Some fonts are too big and some are too small

- The algorithm blocks could be re-organsed to be clear enough to readers. The current algorithm blocks look more sparse and a bit confusing.

- Figures are in quite low resolution, e.g., Fig. 8, 13 is hard to read.

 

I would suggest a major revision and review again.

Comments on the Quality of English Language

Could be improved.

Author Response

Please note that I have used individual colors in the manuscript text to highlight some of the important or specific changes related to your feedback. The responses related to Reviewer 1 are highlighted in red, the responses related to Reviewer 2 are highlighted in green, and the responses related to Reviewer 3 are highlighted in purple. Some of the text has mixed colors to indicate that it relates to feedback from multiple reviewers.

Strengths:

+ the paper is overall technical sound

+ nice visualisations presented in the experimental section

+ some interesting comparisons and analysis presented in the paper

Response:

Thank you, your encouraging comments are invaluable to us, and we are committed to further enhancing the quality and clarity of our research.

Weaknesses:

- One of the major issues with this submission is its novelty. The paper is more like exploring

a combination of existing machine learning methods for the automation of road safety

inspection.

Response:

Thank you for your feedback. While our work does build upon existing machine learning methods, we believe our contributions offer significant advancements in the field of road safety inspection. To address your concerns, we have included the novelty and innovation point in the introduction section (Line 95 to 112):

The study developed a hybrid machine learning system for real-time, privacy-preserving anomaly detection in road safety inspections. The research's contributions are listed as follows:

 

• Produces a traffic safety analysis artefact scalable to scenarios in over 20+ countries and hundreds of similar traffic scenes [10] that employ movable concrete barriers.
• Introduces a semi-automated synthetic data generation method using a novel background cloning technique. The novel approach addresses data sparsity and enhances model training, with repeatability value for other computer vision case studies facing dataset balancing issues.
• Refines classification methods to balance false positives and negatives, improving detection accuracy from 82.6% (reported in earlier peer-reviewed research [11]) to 93.6%. Achieving this 11% increase in accuracy within complex traffic scenes characterized by chaotic backgrounds and lighting conditions underscores the artefact's viability for real-world applications.
• Successfully navigates hazardous traffic scenes for data collection by adhering to industry safety protocols. This repeatable approach provides a comprehensive blueprint for managing similar scenarios, ensuring stakeholder satisfaction and achieving sufficient data.

 

- The paper reviewed some related works, e.g., advances in road defect and anomaly detection;

however, the comparison regarding to existing methods are not analysed and discussed in the

related work section.

Response:

Thank you for your valuable feedback. While our research builds upon and improves previous peer-reviewed work, we have also conducted a secondary comparison to provide a thorough analysis of existing methods. To address your concern, we have made the following additions:

1. Comparative Analysis:
• We have added a comparative analysis in the form of a table and an ROC graph to the manuscript. This comparison includes key performance metrics such as accuracy, precision, recall, and ROC-AUC for our proposed method and other popular algorithms.
2. Enhanced Discussion:
• The discussion section now clearly highlights how our approach outperforms existing methods, emphasizing the improvements and innovations introduced by our research.

These changes ensure that our manuscript not only reviews related works but also provides a detailed comparison with existing methods, demonstrating the effectiveness and advancements of our proposed approach.

- The method section should detail the algorithm proposed, in a clear way. The current paper,

presents some equations, but the equations suffer from several issues, e.g., typos and issues

with index, detailing below.

Response: Thank you for your valuable feedback.

To address this comment, we have made the following additions to the manuscript:

1. Detailed Algorithm Description:

• We have provided a clear and detailed description of the proposed algorithm, outlining each step of the process systematically.
• This includes a step-by-step explanation of the data collection, synthetic data generation, model training, and evaluation processes.
1. Correction of Equations:
• All equations have been thoroughly reviewed and corrected to address any typos and indexing issues.
• We have ensured that standard mathematical notations are used consistently throughout the section.
2. Notation Section:
• Added a notation section that clearly explains the symbols and mathematical notations used in the paper, enhancing clarity, and understanding.

These revisions ensure that the method section now provides a clear and comprehensive explanation of the proposed algorithm, with corrected and accurately presented equations.

- The proposed method is not compared to existing methods, or the variant of the proposed

model. There should be some variants study and ablation study to show the benefits of the

model.

Response: Thank you, to address this we have added a comparative analysis in the form of a table and an ROC graph to the manuscript. This comparison includes key performance metrics such as accuracy, precision, recall, and ROC-AUC for our proposed method and other popular algorithms.

- The discussion section should be presented in a better way rather than listed in bullet points.

The discussion section is also not clear and thorough, especially the discussions on limitations

and future work.

1. Comparative Analysis:
• We have added a comparative analysis in the form of a table and an ROC graph that includes key performance metrics such as accuracy, precision, recall, and ROC-AUC. This comparison highlights the performance of our proposed method relative to several popular algorithms (Section 3.6, Line 557 to 582).
2. Variants Study:
• We have included a study of different variants of our proposed model, exploring various configurations and parameters to demonstrate the benefits and robustness of our approach.
• This section details the performance impact of different network architectures, optimization techniques, and data augmentation strategies.
3. Ablation Study:
• An ablation study has been added to systematically evaluate the contributions of each component of our model. By selectively removing or modifying components, we show the significance and effectiveness of each part of the proposed system (eg, Figure 13 learning rate selection criterion).
• This study helps to illustrate the individual and combined benefits of the various features and techniques incorporated in our model (selecting hybrid features and kernal manipulation).

These additions ensure that our manuscript now provides a thorough comparison with existing. We have updated the relevant sections and also would like to point out where already discussed the failure cases and limitations in multiple sections of the manuscript. Here are the key points:

Background Noise:

Detection accuracy is affected by background noise, making it erratic despite overall improvements.

Reference: Discussion on the need for a more diverse training dataset and how detection accuracy varies under different conditions.

Data from Different Angles:

The necessity for data from various angles is emphasized to further improve the model.

Reference: Mentioned in the context of enhancing robustness and accuracy by expanding data collection protocols.

Detailed Evaluation:

The evaluation of the STENet model includes configurations and performance metrics, highlighting the efficiency and challenges encountered.

Reference: Detailed in the performance evaluation section, comparing SGDM and Adam optimizers, and showing RMSE and validation losses (Table 9 and Table 10).

Robust Performance with Limitations:

While the system performs well in ideal conditions, accuracy varies with different fields of view and lighting conditions.

Reference: Discussed the limitations in terms of dataset diversity and manual frame creation challenges (line 627 to 632, 687 to 693,  709 to 711 and 823).

- Table captions also not clear enough. Table 9 caption does not clearly show the information

presented the paper with a proper explanation of the each column, etc.

Response: Thank you for your feedback, The table captions are updated again for more clarity.

- The evaluations are also quite limited. The authors claim that the proposed method

addresses the issues of detection, but the detections regarding to multiple factors / defects are

not clearly presented.

Response: Thank you for your feedback. We have updated the manuscript to include a comprehensive evaluation of the proposed method, addressing detections across multiple factors and defects. Detailed performance metrics and results are now clearly presented in the revised sections, ensuring a thorough demonstration of our method's effectiveness.

- Some figures have issues with layouts, it would be better to rearrange them to make them look nicer.

Response: Thank you for your feedback. We have addressed the layout issues with the figures in the manuscript. Specifically, Figures 8 and 13 have been rearranged and resized to improve their visual appeal and readability. These changes ensure that the layout is cleaner and more professional.

Thank you for your constructive feedback, which has helped us enhance the presentation quality of our manuscript.

- Eq. (2) seems to have issues with index, and maths symbols. The authors should use standard

maths expressions, e.g., subscript or superscript etc. Issues also inside Eq. (4). The authors are

encouraged to update all the maths equations to make them correct.

Response: Thank you, mentioned equations are now updated (Line 204 and 225).

- It is suggested to have a notation section detailing the maths symbols used in the paper, e.g., regular fonts for numbers, lowercase bold face for vectors, and uppercase bold face for matrices etc.

Response: Thank you, The notation section is added to improve clarity.

- Tables could be rearranged nicely. Some fonts are too big and some are too small

Response: Thank you. Most of the tables have been reformatted, and some have been rearranged to improve the flow of information.

- The algorithm blocks could be re-organsed to be clear enough to readers. The current algorithm

blocks look more sparse and a bit confusing.

 

- Figures are in quite low resolution, e.g., Fig. 8, 13 is hard to read.

Response: Thank you. The figures are now improved for clarity

I would suggest a major revision and review again.

Comments on the Quality of English Language

Could be improved.

Response: Thank you. Another proofreading exercise was done to improve consistency.

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The paper has significantly improved. However, one last concern/suggestion for the authors is to add a graphical abstract or general pipeline to their article. This graphical abstract should summarize the entire article in one concise, pictorial, and visual figure, effectively highlighting the main findings.

Author Response

Round 2:

The paper has significantly improved. However, one last concern/suggestion for the authors is to add a graphical abstract or general pipeline to their article. This graphical abstract should summarise the entire article in one concise, pictorial, and visual figure, effectively highlighting the main findings.

Response:

Thank you for your positive feedback and for recognising the improvements made in our manuscript. We appreciate your suggestion to include a graphical abstract; however, After careful consideration, our team decided against including a graphical abstract, as our paper is already graphics-intensive. We believe adding another visual element might overwhelm the readers and detract from the clarity of the existing figures.

However, in response to your suggestion and to enhance the clarity and organisation of our paper, we have added the following text at the beginning of the Introduction section to outline the structure and content of the manuscript:

"This paper is organised as follows: Section 1 presents an overview of the problem and the research question; Section 2 includes a comprehensive literature review on Automated Road Defect and Anomaly Detection; Section 3 proposes a new methodology based on a hybrid deep learning solution to augment visual inspections using computer vision technology."

We hope this addition provides a clear and concise overview of the paper's structure, helping readers navigate the content more effectively.

Thank you again for your valuable feedback and support.

Reviewer 3 Report

Comments and Suggestions for Authors

After the revision, the paper has improved a lot.

I noticed that there are some issues with the maths expressions, the authors should prepare a minor revision on those.

- Eq. (2), x_i, w_i rather than xi, wi

- Eq. (5), bracket is not in pair

- What is Eq. (10)?  The authors should explain it clearly and whether it is correct or not.

- Fig. 12 (B) is in quite low resolution and the font sizes are small

For the figures, the authors are encouraged to make those figures nicely presented -- the current layout looks not that professional, e.g., some figures could be smaller a bit. Please do work on those.

Comments on the Quality of English Language

Could be improved, e.g., a proof reading of the final version.

Author Response

Round 2:

After the revision, the paper has improved a lot.

I noticed that there are some issues with the maths expressions, the authors should prepare a minor revision on those.

Response:

Thank you for your positive feedback on the improvements made to our paper. We have reviewed and corrected the mathematical expressions as suggested, ensuring they are accurate and clearly presented. We appreciate your guidance and are pleased to have addressed these final details.

- Eq. (2), x_i, w_i rather than xi, wi

Response:

We have corrected the notation in equation (2) to use the appropriate subscript format, changing xi and wi to ​ and ​. This adjustment ensures clarity when representing the pixel locations and intensities in the equation. Thank you for pointing this out.

- Eq. (5), bracket is not in pair

Response:

We have addressed the issue with equation (5) by ensuring all brackets are correctly paired. This correction improves the clarity and accuracy of the equation, making it easier to understand the relationship between the variables. Thank you for highlighting this issue.

- What is Eq. (10)?  The authors should explain it clearly and whether it is correct or not.

Response:

We have provided a clearer explanation of equation (10) within the context of the regionprops function. Equation (10) defines the calculation of the area of a region in an image, representing the total number of pixels within that region. Each pixel contributes a value of 1, effectively counting the pixels to determine the region's area. This explanation clarifies the purpose and correctness of the equation in the context of our methodology. Thank you for your feedback, which has helped us improve the clarity of our presentation.

 

 

- Fig. 12 (B) is in quite low resolution and the font sizes are small

Response:

We have improved the resolution of Fig. 12 (B) to the best extent possible. Unfortunately, due to a data loss on our university-provided computing system, we cannot provide a clearer version of the (B) part of the figure. We appreciate your understanding and have ensured that the current version maintains as much clarity as feasible. Thank you for your feedback.

 

For the figures, the authors are encouraged to make those figures nicely presented -- the current layout looks not that professional, e.g., some figures could be smaller a bit. Please do work on those.

Response:

We appreciate your feedback regarding the presentation of the figures. We have revised the layout and presentation to the best of our capacity, resizing some figures for better visual balance and clarity. Thank you for your constructive comments.

 

Round 3

Reviewer 2 Report

Comments and Suggestions for Authors

None