Automatic and Efficient Detection of Loess Landslides Based on Deep Learning

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

In this paper, deep learning techniques are used to automatically and efficiently detect loess landslides. Coordinate Attention (CA) is integrated into the backbone with the aid of the YOLO model to improve the accuracy on obtaining precise location information and remote spatial interaction data from landslide images. I have several suggestions and comments.

1. The paper writes a very detailed introduction. However, the third paragraph of the introduction can be shortened and more focused on relatively important previous achievement.

2. In Section 2.1, how do you know if the landslides truly occur from the remote images? Do you compare any previous datasets?

3. In Section 2.3, is the Coordinate Attention (CA) firstly proposed by the authors? Or from other previous studies. It is not clear. If it was proposed by others, you need to put references here.

4. In Figure 6, the image processed by the Coordinate Attention module becomes blurring. How can it improve the prediction accuracy?

5. In Figure 10, it is unnecessary to put the original YOLO Head.

6. Line 434, do you use FPS in your equation? I don’t see it. The FPS is used in the table. You should put the explanation of FPS below the table.

7. In the entire manuscript, it should be “Equations” not “Formulas”.

8. Conclusion section should be extended to pros/cons, novelty, future works.

Author Response

Dear Reviewer,

Thank you for giving us the opportunity to submit a revised version of our manuscript "Automatic and efficient detection of loess landslides based on deep learning" for publication in " sustainability" journal. We appreciate your time and effort in providing feedback on our manuscript and thank you for your insightful comments and valuable improvements to our paper. We have incorporated most of the suggestions you made. These changes are highlighted in the original manuscript. Please see the word document below, which is a point-by-point response to the reviewers' comments and concerns.

Sincerely,

The Authors

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The article focused on the development and application of an innovative landslide detection model for monitoring and managing landslide hazards on the Loess Plateau in northwestern China. The authors created a comprehensive dataset consisting of 11,010 images of landslide occurrences on the Loess Plateau. This extensive dataset likely serves as a foundation for training and evaluating their landslide detection model. The use of Coordinate Attention, integrated into the backbone with the YOLO model, aims to capture precise location information and spatial interactions from landslide images. This suggests a focus on improving the accuracy of location data in landslide detection. The inclusion of CBAM in the model's neck is designed to prioritize legitimate landslide objectives while filtering out background noise. This helps extract valid feature information, enhancing the model's ability to focus on relevant landslide areas. The authors introduce a lightweight component called the Decoupled Head, which efficiently extracts classification and location details from landslide images without significantly increasing model parameters. This is intended to improve detection accuracy.The article reports that the improved model achieved landslide object detection at multiple scales with a mean average precision (mAP) of 92.28%, representing a 4.01% improvement compared to the unimproved model.

Unfortunately, the article also has weaknesses. Here are some comments.

Weaknesses of the article:

Ø  The article does not include an analysis of the limitations of its approach or potential risks associated with the proposed model. It would be beneficial to consider situations where the model may not perform effectively.

Ø  The article does not provide detailed information about the dataset itself, such as data collection methods, the representativeness of the dataset, or its public availability. This is crucial for understanding the overall quality of the research.

Ø  The article concentrates on research related to the Loess Plateau in China, which may limit the general applicability of the proposed model to other geographic areas with different conditions.

Ø  The article does not discuss aspects related to the training time of the proposed model and its computational complexity. These aspects are important, especially in the context of practical model deployment.

Strengths of the article:

Ø  The article presents extensive research, including the construction of a large database of landslide images, a comparison of different object detection models, and ablative experiments to assess the effectiveness of individual model components.

Ø  The authors propose an innovative landslide detection model, utilizing various advanced techniques such as Coordinate Attention, Convolutional Block Attention Module (CBAM), Decoupled Head Mini, and the SIoU loss function. This model achieves higher detection accuracy compared to other advanced models, as confirmed by the comparison results.

Ø  The article provides the results of experiments, including the comparison of different object detection models, ablative results of model components, and changes in model performance during training. This allows the reader to understand the effectiveness of the proposed approach.

Ø  The article concludes with detailed findings, summarizing the achievements of their work, highlighting the effectiveness of specific model improvements, and indicating potential applications.

 

About Disscusion chapter

I'm missing the discussion section. I propose to develop it based on the following proposals.

It should delve into a thorough comparison with other models, both advanced and classical, highlighting the reasons behind the superior performance of the proposed model in terms of precision, recall, mAP, and FPS. Additionally, the discussion should focus on the impact of various model enhancements, such as Coordinate Attention, C3CBAM, Decoupled Head Mini, and the SIoU loss function, and how these contribute to the overall effectiveness of the model. An essential topic for discussion is the significance of performance indicators like mean Average Precision (mAP) and Frames Per Second (FPS) in the context of landslide detection. The section should elaborate on why higher values of mAP and FPS are desirable and explore potential practical applications of the model in real-world scenarios, particularly in the context of landslide monitoring. Furthermore, the stability of the model during training, as indicated by achieving stable results on the validation set, should be discussed. The authors may want to provide insights into why model stability is crucial and how it directly influences the quality of landslide detection. The "Discussion" section should also critically analyze the results of ablative experiments, shedding light on the impact of individual model components (CA, C3CBAM, Decoupled Head Mini, SIoU) on its overall effectiveness. A critical aspect of the discussion involves considering the practical implications of the model's findings. This could include identifying potential applications, benefits, and areas for improvement. Additionally, the authors may suggest future research directions and areas that require further exploration and development to advance the field of landslide detection. Ultimately, the "Discussion" section is an opportunity for the authors to provide a nuanced interpretation of their  results, compare findings with existing literature, and offer insights into the broader implications of their work in the field of landslide detection

 

About Conclusion chapter

Consider introducing the "Conclusions" section with a brief summary statement that encapsulates the main findings of the study. This could serve as an initial guide for the reader. Enhance the final paragraph with a concise summary statement that emphasizes the overarching achievement of the study. This statement should encapsulate the key contributions and improvements made by the proposed model.Integrate key metrics, such as precision, recall, and mAP, into the concluding remarks to provide a more comprehensive overview of the model's performance. Specify how these metrics contribute to the overall success of the proposed model. Expand the discussion on the broader implications of the research. Discuss how the enhanced landslide detection model could impact disaster management, remote sensing applications, or related fields. Conclude with a forward-looking section that discusses potential future directions for research in landslide detection. Address any remaining challenges or areas that could be explored to further refine and advance the proposed model.

Here's a revised version that incorporates these suggestions:

 

"In conclusion, our study presents a lightweight network model for landslide recognition in disaster protection, built upon the YOLOv5 framework, Coordinate Attention, C3CBAM, SIoU loss function of YOLOv6, and Decoupled Head Mini. The proposed model effectively identifies multi-scale and small-object landslide features, achieving a notable mean Average Precision (mAP) of 92.28%, showcasing a significant improvement of 4.01% compared to YOLOv5. This research contributes to the ongoing trend in developing network models for landslide object detection, emphasizing the importance of lightweight models that balance accuracy and computational efficiency.

Adding Coordinate Attention enhances the robustness of the YOLO model, improving feature extraction from remote sensing landslide images and subsequently enhancing detection accuracy. The fusion of Neck's C3 module with CBAM further augments the model's capability to identify landslide-specific attributes and distinguish them from background noise. The incorporation of a lightweight feature enhancement module and a feature extraction model addresses the challenge of maintaining accuracy while reducing model volume, making the enhanced landslide detection model well-suited for monitoring and managing landslides. While the decoupled head enhances object detection model accuracy, it comes at the expense of increased computational power. Striking a balance between accuracy and computational effort is crucial, with adjustments such as a 1x1 dimensionality reduction operation being typical before decoupling. Recognizing the unique angles and directional features of remote sensing images of landslides, traditional object detection models may fall short. The introduction of SIoU, considering the vector angle between the ground truth box and the prediction box, proves beneficial in reducing deviation and improving the recognition performance of landslide images in object detection. Looking forward, future research in landslide detection could explore additional challenges and advancements, building upon the insights gained from this study. The developed model holds promise for applications in disaster management, remote sensing, and related domains, contributing to the ongoing efforts in enhancing our understanding and mitigation of landslide risks."

 

Figures 1 must to be improved.

References appear to be optimal.

Comments for author File: Comments.pdf

Author Response

Dear Reviewer,

Thank you for giving us the opportunity to submit a revised version of our manuscript "Automatic and efficient detection of loess landslides based on deep learning" for publication in " sustainability" journal. We appreciate your time and effort in providing feedback on our manuscript and thank you for your insightful comments and valuable improvements to our paper. We have incorporated most of the suggestions you made. These changes are highlighted in the original manuscript. Please see the word document below, which is a point-by-point response to the reviewers' comments and concerns.

Sincerely,

The Authors

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

In the context of landslide detection on the Loess Plateau, the article has made commendable efforts in innovating the detection model, and the results appear promising. However, I still have several queries and suggestions:

1. The utilization of only the RGB values from Google imagery and the exclusion of DEM raise questions. What is the rationale behind not incorporating DEM data? Given the complex surface morphology of landslides, the detection might lack persuasiveness without the assistance of elevation information.

2. How do you ascertain that the detected areas are unequivocally landslides and not merely exposed ground or other possibilities?

3. The division ratio of the training and testing sets is discussed in lines 173-174, yet this ratio does not seem to be depicted in Figure 2. Can this information be visually represented for clarity?

4.  It is recommended to enhance the aesthetic quality of Figure 1, paying attention to cartographic standards and ensuring clear indicators.

Comments on the Quality of English Language

Moderate editing of English language required.

Author Response

Dear Reviewer,

Thank you for giving us the opportunity to submit a revised version of our manuscript "Automatic and efficient detection of loess landslides based on deep learning" for publication in " sustainability" journal. We appreciate your time and effort in providing feedback on our manuscript and thank you for your insightful comments and valuable improvements to our paper. We have incorporated most of the suggestions you made. These changes are highlighted in the original manuscript. Please see the word document below, which is a point-by-point response to the reviewers' comments and concerns.

Sincerely,

The Authors

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

I  am satisfied with the revised manuscript. The paper can be accepted.

Reviewer 2 Report

Comments and Suggestions for Authors The manuscript has been revised in line with the comments made in the review. In view of this, I have no further comments on the manuscript. The manuscript can be published in its accepted form.