Automatic Defect Detection of Pavement Diseases

Round 1

Reviewer 1 Report

Overall Decision：Major revision

This manuscript introduces a novel deep convolution neural network--DASNet, which can be used to identify road diseases. The network employs deformable convolution instead of regular convolution as the feature pyramid's input, adds the same supervision signal to the multi-scale features before feature fusion, decreases the semantic difference, extracts context information by residual feature enhancement, and reduces the information loss of the pyramid's top-level feature map. Considering the unique shape of road diseases, it is easy to cause an imbalance problem between the foreground and background and introduces the sample weighted loss function. In summary, the research is interesting and provides valuable results, but the current document has several weaknesses that must be strengthened in order to obtain a documentary result that is equal to the value of the publication.

(1) At the thematic level, the proposal provides a very interesting vision, as pavement diseases detection is an important task to ensure road safety，However, due to the irregular shape and large-scale differences in road diseases, and the imbalance between the foreground and background, the task is challenging. This manuscript proposes an innovation in the detection accuracy of deep learning network and designs a verification test, which is worthy of affirmation in terms of research methods. However, there are some problems in this paper, such as unclear background research and lack of verification of real scenes, which lead to the lack of practical value of the research results.

(2) The network structure proposed in this paper is relatively complex, which improves the detection accuracy of the model. However, I have some doubts about whether the detection speed of the model can meet the actual demand.

(3) The abstract is complete and well-structured and explains the contents of the document very well. Nonetheless, the part relating to the results could provide numerical indicators obtained in the research.

(4) The number of the keywords is a little large and the coverage is too wide, which leads to the main contribution of the research being covered.

(5) The author should review the road disease detection methods based on deep learning in recent years combined with the paper data.

(6) It is suggested that the author provide some background investigations related to practical application requirements, so as to facilitate the verification of the practical value of the method combined with the experimental results in this paper (Visual measurement of dam concrete cracks based on U-net and improved thinning algorithm, Journal of Experimental Mechanics. Seismic performance evaluation of recycled aggregate concrete-filled steel tubular columns with field strain detected via a novel mark-free vision method, Structures).

(7) It is suggested to omit unnecessary theoretical explanation and focus on the contribution of this research and the problems solved.

(8) In the section describing the network structure (chapter 2), the author should reduce the length of the text introduction.

(9)  Chapter 3: Road Diseases Detection Model—DASNet shows the main research innovations and work processes of this study. I wonder how much the complexity of the model affects the detection speed.

(10)  4.2: Too much space is devoted to the introduction of various evaluation indicators in this paper.

(11)  4.4: The detection speed is also an important index for practical application.

(12)  4.6: The YOLO series model has been updated many times. Why did the author choose YOLOV3?

(13) In the conclusion part, the author did not summarize the shortcomings of this study and the directions for future research.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper focuses on the automatic detection of road pavement distresses. This is a high topic of research in the domain of transport infrastructures. The paper is well-written and structured. Some comments are recommended to improve the paper.
The abstract says: “Pavement diseases detection is an important task to ensure road safety”. I agree with the authors, but pavement disease detection is also crucial for pavement maintenance. The authors also said: “Manual visual detection requires a significant amount of time and effort”. And about costs?
It is unclear if the detection tool described in the paper applies to asphalt or concrete pavements.
Besides the detection of the distresses, another important task is the severity influence on the detection method. Does the distress severity influence the proposed method for automatic detection? Any impact on the accuracy of the methodology? All the pavement diseases could be identified by the proposed tool?
Please ensure all acronyms are introduced in the manuscript (e.g., RCNN).
Examples of the images in Figure 1 (diseases) should be explained in more detail in the text. The caption of this figure is not clear.
Please, correct the title of section 3 (Road Diseases Detection Model – DASNet)
Information about the type of pavement (asphalt or concrete) is missing. Description of the diseases observed in each image of Figure 8.
Please, complete the conclusions with some weak aspects of the proposed methodology and further research on the paper's topic.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

ACCEPT