All reviews of published articles are made public. This includes manuscript files, peer review comments, author rebuttals and revised materials. Note: This was optional for articles submitted before 13 February 2023.
Peer reviewers are encouraged (but not required) to provide their names to the authors when submitting their peer review. If they agree to provide their name, then their personal profile page will reflect a public acknowledgment that they performed a review (even if the article is rejected). If the article is accepted, then reviewers who provided their name will be associated with the article itself.
The paper is ready to be accepted.
[# PeerJ Staff Note - this decision was reviewed and approved by Xiangjie Kong, a PeerJ Section Editor covering this Section #]
No comment.
No comment.
No comment.
I am happy to recommend acceptance, having seen this evolve into a well contained research article with clear future plans. Very commendable.
The recommendations from the two reviewers are divergent. Please revise the paper following all the comments from the reviewers.
The introduction should be compared with other Geotechnical Damage Detection methods based on deep learning. The novelty of the paper lies solely in performing a binary classification task using a 3-layer CNN. Why not employ established CNN models such as VGG, ResNet, and others?
The experimental section should include experiments comparing the proposed method with some mainstream object detection approaches.
The experimental section should include the experimental results of author's method on publicly available datasets relevant to Geotechnical Damage Detection, rather than conducting experiments solely on author's dataset.
The dataset for the paper comprises only 674 images, which is insufficient to simulate real-world scenarios or obtain meaningful experimental data.
The experimental section provides results in terms of a confusion matrix and ROC curve. The experimental part should include comparative tests between author's method and other object detection approaches. Additionally, applying author's method to experiments on other publicly available datasets is crucial to thoroughly demonstrate its effectiveness.
This paper introduces a simple neural network model for Geotechnical Damage Detection and proposes a dataset. However, there is insufficient evidence in the paper to validate the effectiveness of both the model and the dataset. Additionally, the current workload in the paper is insufficient to support the arguments they put forward.
The paper has been rewritten almost entirely, which has improved the flow considerably. The authors have made a commendable effort to share data and code, the paper is now also reasonably self contained. The title change was very timely.
The design has not changed, but it has been better motivated. I believe there is sufficient effort here.
The results of the network described are good, and they seem robust to a certain amount of weather "noise". However, I remain unconvinced about the model's performance on images under different weather conditions, e.g. in a practical situation, one would expect pouring rain to accompany a mudslide. This is almost the equivalent of adding noise to the images, and neural networks are notorious sensitive to noise. I would suggest rewriting the conclusions. Although the results are promising from a computer science perspective, the applicability in "live" applied situations is not wholly discussed and should be rephrased.
I am very impressed by the improvement made. It is commendable.
Based on the comments and recommendations from the reviewers, the paper needs essential revision and improvement.
**PeerJ Staff Note:** Please ensure that all review, editorial, and staff comments are addressed in a response letter and any edits or clarifications mentioned in the letter are also inserted into the revised manuscript where appropriate.
no comment
no comment
The method uses a customized CNN-based architecture with five convolutional-pooling layers and two fully connected layers to classify the damaged or intact image. How about the performance of the tranditional method ,like SVM or Random Forest?How about the deeper CNN layers?
The authors have taken pains to describe the existing literature on which they have built their binary classifier. However, there are gaps in terms of the raw data being shared, crucially the code used has not been provided in the supplementary either. Raw data should have also been made available on Figshare or some other dataset sharing platform. As written, it is not self contained, and there seems to be an over-reliance on existing research, for example, even the ADAM optimizer and its equations are omitted.
There are no standard training / test loss plots to judge over-fitting from. There are no baseline models nor is there enough context on if a binary classifier is useful to geotechnical damage and detection, especially since the models have no predictive ability in the wider context of geotechnical damage and detection. To "detect threats" is mentioned as the goal, but a binary classifier does not provide much in the way of preventing accidents. To improve this, consider linking this to a phenomological model of damage and trying to interpret the probabilities as degrees of damage.
As discussed above, the data is not provided, nor is the code, and the results / benefits are unclear. Additionally, the models do not have any preprocessing for time / weather / photo conditions etc.
Unfortunately at this level of rigor it is not possible to amend this work without a full rewrite, but I am convinced with effort, this will be useful and (most importantly) correct and reproducible research paper.
All text and materials provided via this peer-review history page are made available under a Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.