A Complete System for Automated Semantic–Geometric Mapping of Corrosion in Industrial Environments
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis manuscript presents a portable system for automated detection and mapping of corrosion in industrial environments by integrating LiDAR-based SLAM and deep learning-based semantic segmentation. The approach aims to address limitations of traditional corrosion detection methods by offering a non-intrusive, low-cost solution with high portability.
The paper demonstrates good technical foundation in both localization and semantic segmentation. However, several significant limitations need addressing, particularly the absence of a clear definition of "corrosion" and insufficient validation in diverse environments. The innovation in LiDAR-based mapping techniques appears limited as these are well-established technologies.
1. Section 2.2: Provide a clear definition of corrosion in the context of this study, including visual characteristics used for identification. Discuss how the system differentiates between actual corrosion and simple discoloration or stains.
2. Section 4.5: Include analysis of semantic segmentation performance under varying environmental conditions such as lighting. The current evaluation lacks validation in diverse settings.
3. Section 4.6: Enhance the outdoor experimental validation with quantitative metrics similar to those provided for indoor experiments. The current outdoor validation seems primarily qualitative.
4. Section 4.3-4.4: Provide more information about the dataset, including different types of corrosion (uniform corrosion, pitting, etc.) and metal surfaces. Discuss the quality and potential limitations of the manual annotation process.
5. Section 5: Expand the discussion on practical limitations of the system for industrial deployment, addressing challenges of false positives and their impact on system reliability.
* Language and Etc.
1. Page 2, line 38-39: "...may lead to lower insurance costs, as the risk of failures and subsequent claims is reduced." The sentence could be improved for clarity.
2. Page 18, line 18: "...with less then 0.05m..." should be "...with less than 0.05m..."
3. The use of conversational phrases such as 'please see Table 2' (p.18) is inappropriate in formal academic writing. Replace with more objective expressions like 'as shown in Table 2' or 'refer to Table 2'."
4. Table 2 (Page 19): The dataset splitting ratios don't add up correctly - the validation and test sets are both listed as 10%, but the total percentages sum to 100%. This suggests an error in either the percentages or the actual number of images used. Images for train accounts for 60% of total image sets??
Author Response
Please see the attached rebuttal pdf file
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for Authors1.In this paper, it is interested that the automated semantic-geometric mapping of corrosion in industrial environments is developed. However, the main contributions and their advantages must be specified.
2.In the image-based corrosion identification, it is not clear how to achieve high accuracy with limited training data.
3.Comparison results with other mapping methods should be provided.
4.What is the advantage for the introduced semantic-geometric mapping, and what is the difference?
5.Tables 1 and 2 should be formatted uniformly (e.g., the unit "s" should be clearly defined as "seconds"), and the datasets source should be supplemented.
Comments on the Quality of English LanguageN/A
Author Response
Please see the attached rebuttal file
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThis paper propose a complete system for semi-automated corrosion identification and mapping in industrial environments. Overall, the structure is good following and presenting the authors findings accurately. However, the manuscript still needs major improvements before it is ready for publication. My specific comments are shown as follows
1、What is the localization accuracy in outdoor environments?
2、This paper lacks an evaluation of the mapping quality.
3、The goal of this paper is to build semantic-geometric maps of industrial environments. However, this paper neither presents nor evaluates semantic maps.
Author Response
Please see the attached rebuttal file
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsThe comparison results need to be given to show achieving high accuracy with limited training data.
Author Response
We thank the reviewer for the comments that helped us improved the paper. Unfortunately, since the scope of the paper was on the overall 3D corrosion mapping architecture, and since we run out of resources for the project. More experiments is no longer possible. We note however, that in a previous publisher work [1] we focused on evaluating our corrosion segmentation dataset as well as different models. However, to make this clear we added the following paragraph in the experiments section:
"Although a detailed comparative evaluation with varying training sizes is beyond the scope of this work, we refer the interested reader to our previous study dedicated to corrosion segmentation \cite{visionCorrosionIST2023}, where we thoroughly evaluated multiple state-of-the-art models on the same dataset under different training conditions."
[1] de Figueiredo, Rui Pimentel, and Simon Bøgh. "Vision-Based Corrosion Identification Using Data-Driven Semantic Segmentation Techniques." 2023 IEEE International Conference on Imaging Systems and Techniques (IST). IEEE, 2023.
Reviewer 3 Report
Comments and Suggestions for AuthorsThis paper can be accepted.
Author Response
We thank the reviewer once again for the comments that helped us significantly improve the quality of the paper.