A Novel YOLOv10-DECA Model for Real-Time Detection of Concrete Cracks

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

It is a well written article. Well done.

Author Response

Comments 1: It is a well written article. Well done.

Response 1: Thank you very much for the strong support of our work. We are truly grateful for your positive feedback. We deeply appreciate your kind words and hope that our article contributes valuable insights to the field.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

General Comments:

The paper presents an innovative approach to concrete crack detection using the YOLOv10-DECA model. The integration of the Dynamic Efficient Channel Attention (DECA) module with the YOLOv10 architecture shows promising improvements in accuracy and speed, making it well-suited for real-time applications. The paper is well-structured and provides a comprehensive overview of the methods and results. However, there are a few areas that could benefit from further clarification and expansion.

 

Specific comments

The introduction briefly mentions existing methods but fails to thoroughly analyze their limitations, particularly in handling complex environments and real-time constraints.

The paper provides insufficient technical details on the specific advantages of the DECA module over the ECA, such as how it enhances feature extraction and decision-making.

A flowchart or diagram illustrating the methodological framework is missing, which would greatly aid in understanding the integration and flow of the DECA module.

There is limited discussion on the variety within the dataset, including aspects like crack types, environmental conditions, and backgrounds.

The application of data augmentation techniques lacks clarity, specifically whether they were uniformly applied across all images or tailored for specific scenarios.

Performance graphs do not include error bars, which are crucial for assessing the variability and statistical significance of the results.

The results section does not provide confidence intervals, which are necessary to understand the reliability and robustness of the performance metrics.

The paper lacks a comparative analysis with other models, such as the techniques used in Tempnet used in civil engineering, which is essential for contextualizing the improvements offered by the YOLOv10-DECA model.

There is insufficient exploration of specific case studies where the DECA module significantly impacts model performance, limiting insights into practical applications.

The discussion on how the model performs in real-time scenarios is limited, leaving questions about its practical deployment and speed in diverse environments.

The rationale behind the chosen parameter settings is not adequately explained, making it difficult to replicate or adjust the model for different contexts.

The paper does not justify the dataset size used, raising concerns about whether it is representative enough to ensure model generalization.

The paper lacks clarity on why specific evaluation metrics were chosen and how they relate to the goals of crack detection.

There is no information on the training duration or computational resources required, which is critical for assessing the model's efficiency and feasibility.

The benefits of using transfer learning are not fully explained, missing an opportunity to highlight its impact on model performance and training time reduction.

The paper does not sufficiently discuss the limitations of the proposed model, such as potential weaknesses in detecting certain crack types or conditions.

There is no exploration of potential biases within the dataset, which could affect the model's performance and fairness across different scenarios.

The scalability of the approach to larger datasets or different infrastructure types is not addressed, leaving questions about its broader applicability.

The methods used for interpretability analysis are not detailed enough, limiting the understanding of how model decisions are made and validated.

 

The paper does not discuss potential directions for future research, missing an opportunity to suggest improvements or new avenues for exploration in crack detection.

 

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

This paper studies YOLO based deep learning models to detect concrete cracks from images, and proposes a YOLOv10-DECA model that combines YOLOv10 architecture with dynamic efficient channel attention module. The authors conduct comprehensive experiments to compare the accuracy and other major metrics across 4 YOLO based models, and the results show that YOLOv10-DECA model has prominent improvements.

I think the paper is well organized, and the only question is why real-time detection is required, which is not clearly stated in this paper. For example, is there any consequence when a crack is detected in 1 minute other than 1second?

Author Response

Comments 1: This paper studies YOLO based deep learning models to detect concrete cracks from images, and proposes a YOLOv10-DECA model that combines YOLOv10 architecture with dynamic efficient channel attention module. The authors conduct comprehensive experiments to compare the accuracy and other major metrics across 4 YOLO based models, and the results show that YOLOv10-DECA model has prominent improvements.

I think the paper is well organized, and the only question is why real-time detection is required, which is not clearly stated in this paper. For example, is there any consequence when a crack is detected in 1 minute other than 1second?

Responses 1: Thank you very much for your recognition. Regarding the requirement for real-time detection, we provide the following explanation. Our team hopes to deploy this model on drones or patrol robots in the future. In certain specific detection scenarios, such as inspections of tunnels or bridges, it is necessary to process images in real-time to locate the position of cracks. In these scenarios, delays in detection can lead to decreased monitoring efficiency and increased inspection time and costs. Therefore, there are high requirements for the detection speed of the model. To better explain this situation to the readers, we have added the following content to the paper:

“The purpose of real-time detection is to facilitate the deployment of the model onto devices such as drones or patrol robots. In certain specific detection scenarios, such as inspections of tunnels or bridges, the model is required to process images in real-time to quickly locate the positions of cracks. In these scenarios, longer detection delays can lead to decreased monitoring efficiency, increased inspection time, and higher costs. Therefore, the model must have high detection speed.”

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

accept