A One-Step Methodology for Identifying Concrete Pathologies Using Neural Networks—Using YOLO v8 and Dataset Review
, Anselmo Cardoso de Paiva 1, Geraldo Braz Junior 1, João Dallyson Sousa de Almeida 1, Aristófanes Corrêa Silva 1, António Manuel Trigueiros da Silva Cunha 2,3 and Sandra Cristina Alves Pereira da Silva Cunha 2,4,*Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsReview of "A One-Step Methodology for Identifying Concrete Pathologies Using Neural Networks. Using YOLO v8 and Dataset Review”
The paper by Diniz et al. proposes a new one-step methodology for detecting concrete pathologies using neural networks, specifically using YOLO model. This work aims to improve the efficiency and accuracy of identifying concrete defects, which is important for safety in civil structures. The paper is well-written with some suggestions for the authors to further improve the work.
1. The methodology section is well-structured, providing details on the image acquisition, detection, and classification processes. However, the data preprocessing section is unsatisfactory. The paper would benefit from a deeper discussion for the data preprocessing steps. How did the authors process these raw images?
2. The training and testing matrices are missing in the paper. The loss as a function of training epochs should be plotted to ensure no overfitting during training.
3. The authors should list the training and testing time to demonstrate their model efficiency.
4. To demonstrate the practical usage, why don’t the authors use the trained ML model to make predictions on real concrete samples beyond the images in the Concrete Defect Bridge Image Dataset?
5. Some recent seminal papers and reviews about using machine learning to detect cracks and defects can be cited and discussed for readers to better understand the field:
(1) Zhang, E., Dao, M., Karniadakis, G. E., & Suresh, S. (2022). Analyses of internal structures and defects in materials using physics-informed neural networks. Science advances, 8(7), eabk0644.
(2) Jin, H., Zhang, E., & Espinosa, H. D. (2023). Recent advances and applications of machine learning in experimental solid mechanics: A review. Applied Mechanics Reviews, 75(6), 061001.
Author Response
Thank you for your contribution to making the article better!
The response to the questions is attached.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe manuscript proposes a methodology for visually inspecting concrete structures using deep neural networks, with a focus on detecting common pathologies such as cracks, fragmentation, efflorescence, and corrosion stains. The method utilizes YOLO versions 4 and 8 for detection and classification, respectively, and is tested for accuracy using the Ozgnel dataset for classification and the CODEBRIM dataset for detection. The study aims to increase the effectiveness of pathology detection while reducing human errors and analysis time.
The study addresses a practical need for automated detection of concrete pathologies, which is crucial for infrastructure maintenance and safety. The methodology is well-described, detailing the use of YOLO v8 for classification and v4 for detection, along with the datasets employed. This clarity enhances the reproducibility of the study.
Achieving 99.65% accuracy in classification indicates the effectiveness of the proposed methodology. The use of deep neural networks streamlines the detection process, potentially saving time and reducing errors compared to manual analysis.
Provide a more detailed description of the methodology, including the preprocessing steps applied to the images, hyperparameter tuning for the YOLO models, and the training/validation process. This information would help readers understand the nuances of the approach. Besides, introduce more recent studies on the application of lightweight CNN models for detection, such as: https://doi.org/10.1007/s00170-022-10335-8
Include a comprehensive evaluation of the detection performance, including metrics such as precision, recall, and F1 score, to assess the model's effectiveness in identifying concrete pathologies. Discuss potential limitations of the proposed methodology, such as challenges with detecting certain types of pathologies or variations in lighting conditions. Addressing these limitations would provide a more balanced perspective on the applicability of the approach.
The manuscript presents a valuable contribution to the field of infrastructure maintenance by proposing a methodology for automated detection of concrete pathologies using neural networks. The study demonstrates high accuracy in classification and highlights the potential for streamlining pathology detection processes. However, further detail on the methodology, validation metrics, and limitations would enhance the clarity and credibility of the findings. With these improvements, the study would offer significant value to researchers and practitioners in the field of civil engineering and infrastructure management.
Comments on the Quality of English LanguageEnglish language used throughout the manuscript is generally good.
Author Response
Thank you for your contribution to making the article better!
The response to the questions is attached.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThe authors have addressed all my comments properly. The manuscript can be accepted as it is now.
