Semantic Segmentation Network for Unstructured Rural Roads Based on Improved SPPM and Fused Multiscale Features

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Pros:

1. Segmenting unstructured rural roads is an important and relevant topic.
2. The enhancements made to the PP-Liteseg model, particularly with the SP-SPPM and BAM modules, effectively tackle challenges like irregular road shapes and occlusions.
3. The experiments are well-executed, with clear comparisons in section 3 to other models like UNet, ENet, BiSeNetv1, and BiSeNetv2, highlighting the advantages of the improved PP-Liteseg model.
4. Explanation of dataset collection and annotation adds credibility to the study.

Cons:

1. The contribution is more of a incremental advance on existing models rather than introducing completely new concepts.
2. The dataset is based on a specific region (Xinjiang, China). How can authors make it more generalizable so that the results are to other rural areas.
3. The added complexity from the SP-SPPM and BAM modules could make the model difficult to use in real life setup. Authors should discuss the trade-off between accuracy and efficiency.
4. The replication package should be accessible hassle-free. Please address the same in the final version if possible.
5. No threats to validity section. It should cover construct, internal, and external validity.

Comments for Authors:

1. Novelty: Authors should explain how the pape could support new research ideas or be applied in different areas, which would make the uniqueness of the work stand out more.
2. Significance: Authors should discuss on discuss how the model could also work in other possible difficult rural setting.
3. Soundness: The methods you used are overall solid. Please add more discussion on explaining more about why certain models (like BiSeNetv1 or UNet) didn't perform as well in some cases.
4. Presentation: The paper is well-organized overall, but the introduction could do a better job of explaining why rural road segmentation is important beyond just farming uses. Also, Section 3 can be broken in separate sections to improve reading.
5. Other Comments:
• The conclusions could be improved  by giving specific suggestions for future research or how to use the findings in practice.

Author Response

Dear Reviewers:

Thank you for your letter and for the reviewers’ comments concerning our manuscript entitled “ Semantic Segmentation Network for Unstructured Rural Roads Based on Improved SPPM and Fused Multiscale Features” (ID: applsci-3189148). Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our research. We have studied comments carefully and have made corrections which we hope meet with approval.

 

Comments 1: Authors should explain how the paper could support new research ideas or be applied in different areas, which would make the uniqueness of the work stand out more.

Response: The concluding section looks to the future and suggests some possible measures. Ideas for other new research are also provided.

 

Comments 2: Authors should discuss on discuss how the model could also work in other possible difficult rural setting.

Response: Section 5.4 has been added to the experimental section to test the model on the publicly available dataset IDD to validate the possibility of using the model of this paper on other unstructured roads.

 

Comments 3: The methods you used are overall solid. Please add more discussion on explaining more about why certain models (like BiSeNetv1 or UNet) didn't perform as well in some cases.

Response: The comparison experiment section was rewritten to describe and analyze the reasons for the poor performance of the other models.

 

Comments 4: the paper is well-organized overall, but the introduction could do a better job of explaining why rural road segmentation is important beyond just farming uses. Also, Section 3 can be broken in separate sections to improve reading.

Response: The introduction section has been rewritten and the importance of semantic segmentation of rural roads has been explained. Section 3 has also been split to enhance the reading.

 

Comments 5: The conclusions could be improved by giving specific suggestions for future research or how to use the findings in practice.

Response: The conclusions section has been revised to include future work and research that can be applied.

 

The full text was rechecked for logical and grammatical issues, with all changes marked in red.

Once again, thank you very much for your comments and suggestions. We hope that the revised manuscript can be accepted by Applied Sciences.

Thank you and best regards.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The article refers to the study of an improved PP-Liteseg semantic segmentation model proposed, which consists of a bar pool simple pyramid pooling module and a parallel feature fusion module, which enhances the extraction of image features and can achieve better segmentation results. This interesting topic is within the scope of the Agriculture Journal. However, I have a few comments. Please refer to them.

 

1) Is it feasible to apply the proposed PP-Liteseg model to other rural environments beyond those studied?

2) What methodological justification supports the construction of the Rural Roads Dataset (RRD), and how does it ensure its representativeness?

3) What are the main challenges in collecting data for rural roads, and how were they addressed in the study?

4) How is the accuracy of the proposed model evaluated, and what are the key criteria for determining its success in different types of rural scenes?

5) What are this study's key contributions to computer vision and smart agriculture?

6) What are the study's main limitations, and how could they be overcome in future research on rural road segmentation?

Comments on the Quality of English Language

Minor editing of English language required.

Author Response

Dear Reviewers:

Thank you for your letter and for the reviewers’ comments concerning our manuscript entitled “ Semantic Segmentation Network for Unstructured Rural Roads Based on Improved SPPM and Fused Multiscale Features” (ID: applsci-3189148). Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our research. We have studied comments carefully and have made corrections which we hope meet with approval.

 

Comments 1: Is it feasible to apply the proposed PP-Liteseg model to other rural environments beyond those studied?

Response: Section 5.4 has been added to the experimental section to test the model on the publicly available dataset IDD to validate the possibility of using the model of this paper on other unstructured roads.

 

Comments 2: What methodological justification supports the construction of the Rural Roads Dataset (RRD), and how does it ensure its representativeness?

Response: The theory followed in the construction of this dataset is added at the end of the Data Acquisition section. The RRD dataset is constructed with a clear methodological basis, which ensures the representativeness and applicability of the dataset by covering different areas, road types, and diverse obstacles. The quality and comprehensiveness of the dataset are further enhanced by the technical means of data collection and labeling. These designs make RRD not only applicable to the task of semantic segmentation of rural roads but also extend its application to similar complex unstructured scenarios.

 

Comments 3: What are the main challenges in collecting data for rural roads, and how were they addressed in the study?

Response:

 (1)Diverse Road Conditions and Unstructured Environments.

• Challenge: Rural roads often lack consistent structure, with irregular shapes, undefined or fuzzy boundaries, and mixed surfaces like dirt, gravel, asphalt, and non-hardened roads. These conditions can vary greatly across regions and seasons, making it difficult to standardize data collection.
• Solution in the Study: The study addressed this challenge by capturing a wide range of rural road types from multiple regions in Xinjiang, China (e.g., Shawan, Fukang, Nanshan, Urumqi). This geographic diversity ensured that the dataset included various road conditions, from well-maintained asphalt roads to rugged, non-hardened dirt paths, thus providing a representative sample of rural environments.

(2)Presence of Obstacles and Occlusions.

• Challenge: Rural roads often have obstacles like vegetation, fences, animals, agricultural machinery, and poles, making it difficult to collect clear road images and increasing the complexity of the scene for semantic segmentation.
• Solution in the Study: The dataset included various labeled objects such as vegetation, poles, barriers, vehicles, and more. By specifically annotating these objects, the study ensured that the data reflected the true complexity of rural roads and equipped the model to handle occlusions and obstacles commonly encountered in rural environments.

(3)Scarcity of Annotated Rural Road Data.

• Challenge: Unlike urban environments where datasets and pre-labeled data are more readily available, rural road datasets with precise annotations are scarce. This scarcity poses a challenge for developing robust models that can perform well in these environments.
• Solution in the Study: The study constructed a new rural road dataset (RRD) with detailed pixel-level annotations using a custom-built annotation platform (CVAT). This ensured that the data was accurately labeled across a variety of rural road scenes, addressing the lack of existing annotated datasets for this specific context.

 

Comments 4: How is the accuracy of the proposed model evaluated, and what are the key criteria for determining its success in different types of rural scenes?

Response: The accuracy of the proposed model in the study is evaluated through various performance metrics and specific testing on diverse rural scenes. The key criteria for determining its success in different types of rural scenes include the following:

1. Performance Metrics: The evaluation of the model's accuracy relies on several standard metrics used in semantic segmentation tasks. These metrics allow for a comprehensive assessment of the model’s ability to correctly identify and segment rural road features. The key metrics used in this study are Mean Intersection over Union (MIoU), Kappa Coefficient, and Dice Coefficient.
2. Key Criteria for Success in Rural Scenes:(1) Handling Unstructured and Non-Hardened Roads. (2)Distinguishing Between Diverse Obstacles. (3)Segmentation of Small and Irregular Objects. (4)Edge Detection and Boundary Precision.
3. Comparison with Baseline Models: The accuracy of the proposed model is evaluated through a combination of metrics such as MIoU, Kappa coefficient, and Dice coefficient, which collectively assess its ability to handle unstructured roads, obstacles, environmental variations, and small object segmentation. The model’s success in different rural scenes is determined by its robustness in these key areas, with comparisons to baseline models confirming its superior performance in rural road segmentation tasks.

 

Comments 5: What are this study's key contributions to computer vision and smart agriculture?

Response: 1. Contributions to Computer Vision:(1) Improved Semantic Segmentation Model for Unstructured Environments.(2) The Strip Pooling-Simple Pyramid Pooling Module(SP-SPPM). is proposed. (3) The Bottleneck Unified Attention Fusion Module (B-UAFM) is proposed. (4) Creation of a New Rural Roads Dataset (RRD).

2. Contributions to Smart Agriculture:(1) Enhanced Autonomous Navigation for Agricultural Machinery.(2) Improved Path Planning and Task Execution. (3)Promoting Sustainability in Agriculture.

 

Comments 6: What are the study's main limitations, and how could they be overcome in future research on rural road segmentation?

Response: Limitations and future perspectives were added to the conclusion section.

 

The full text was rechecked for logical and grammatical issues, with all changes marked in red.

Once again, thank you very much for your comments and suggestions. We hope that the revised manuscript can be accepted by Applied Sciences.

Thank you and best regards.

Reviewer 3 Report

Comments and Suggestions for Authors

The paper presents an approach for semantic segmentation for the specific case of rural roads. The paper seems to have some interesting contribution, but the presentation and explanation needs to be considerably improved.

The abstract is unclear and unfocused, it should be completely rewriten. The problem and contribution needs to be stated, in addition to the findings. In the description of the method is not clear with are the novel parts proposed by this work. Moreover, since there is not description of related literature, the novelty is even more unclear. A related works sections must be including discussing the limitation of related works and the key aspects of the new proposal. Also, since several baselines are included in the comparison, a description of the main characteristics should be explained in this section. In the report of experiments differences are small in some case, the statistical significance of them have to be also detailed.

References in the text needs to be corrected. For example, this is incorrect “Duong L T[ 12 ]”. Also all references are without a space before [.

 

Comments on the Quality of English Language

The English have some small mistakes and details of presentation.

Author Response

Dear Reviewers:

 

Thank you for your letter and for the reviewers’ comments concerning our manuscript entitled “ Semantic Segmentation Network for Unstructured Rural Roads Based on Improved SPPM and Fused Multiscale Features” (ID: applsci-3189148). Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our research. We have studied comments carefully and have made corrections which we hope meet with approval.

 

Comments 1: The abstract is unclear and unfocused, it should be completely rewritten.The problem and contribution need to be stated, in addition to the findings.

Response: The abstract section was rewritten to make it clearer and more focused.

 

Comments 2:In the description of the method is not clear with are the novel parts proposed by this work. Moreover, since there is no description of related literature, the novelty is even more unclear. A related works section must be including discussion the limitation of related works and the key aspects of the new proposal. Also, since several baselines are included in the comparison, a description of the main characteristics should be explained in this section.

 

Response: A related work chapter is added, the strengths and weaknesses of the baseline model and related work are described, and the innovations of the work are listed at the end of the chapter.

 

Comments 3: In the report of experiments differences are small in some case, the statistical significance of them have to be also detailed.

Response: The section on comparative experiments was rewritten to analyze the reasons for the small gaps in some of the evaluation metrics.

 

Comments 4: References in the text need to be corrected. For example, this is incorrect “Duong L T[ 12 ]”. Also, all references are without a space before [.

Response: References throughout the text have been revised by adding spaces.

 

The full text was rechecked for logical and grammatical issues, with all changes marked in red.

Once again, thank you very much for your comments and suggestions. We hope that the revised manuscript can be accepted by Applied Sciences.

Thank you and best regards.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

I do not have further comments

Comments on the Quality of English Language

Minor editing of English language required.