Black Ice Classification with Hyperspectral Imaging and Deep Learning
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe paper presents an original classification algorithm for black ice detection on the roads. The novelty of the paper is that the authors have used in their research a combination between hyperspectral imagining and a deep learning model.
The paper is well written, clearly, and well organized. The scientific level is good, the results are convincing and strongly argued.
The paper may be accepted after a minor correction. In this sense, please revise the organization of chapter 5.2 “Visualization of future maps”. There are too many subtitles.
Author Response
Dear Reviewer,
Thank you so much for taking time amidst your busy schedule. I appreciate your comments and feedback on our manuscript. I have corrected your mentioned point and modified the manuscript accordingly.
Thank you again for your valuable input on our manuscript which helped us enhance the quality of the paper.
Sincerely,
Chaitali Bhattacharyya
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsIn this manuscript, the author used a 2D-3D Convolution Neural Network to classify hyperspectral images of black ice. Using principal component analysis to preprocess spectral data reduces image cubes' dimensionality and achieves better black ice detection. The experimental results confirm the good performance and accuracy of the proposed method. However, the reviewer has some comments.
1. It is recommended that the references be listed in citation order.
2. In lines 19-21, the sentence "... and the result has been compared with one of the existing methods.". The sentence " The proposed method was then compared with the existing method for better evaluation." has repeated meanings. Please express them concisely.
3. All equations in the article are suggested to be numbered.
4. There are some writing mistakes in lines 212, 383.
5. Please pay attention to the order of image annotation and the standardized use of punctuation in the article.
6. Line 239, “So for a rough idea about black ice, spectral reflectance curves are in Fig. 6, Fig. 7, and Fig. 8 (from 10 meters, 20 meters, and 30 meters away, respectively) …”, Figure 6,7 and 8 do not correspond to this paragraph of text.
7. Figure 14 is incomplete, please modify it.
Author Response
Dear Reviewer,
Thank you so much for taking time amidst your busy schedule and give such detailed feedbacks and suggestions. I appreciate your comments on our manuscript. I have corrected your mentioned point and modified the manuscript accordingly.
Thank you again for your valuable input on our manuscript which helped us enhance the quality of the paper.
Sincerely,
Chaitali Bhattacharyya
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe authors propose using hyperspectral imaging in collaboration with a deep learning model for the classification of black ice on roads. The idea seems very interesting, but the study needs further improvement as some aspects appear to be lacking. Below are the main suggestions and comments that the authors should consider to enhance the quality of the paper:
· The authors should delve deeper into the state of the art, assessing whether the proposed method is superior to existing approaches in the literature.
· In the text, there is not always a reference to figures and/or tables, and explanations for these are often missing.
· The "Methodology" section is generally lacking. The method, neural network, and dataset should be described in more detail, with dedicated subsections.
· Regarding the dataset, the authors should provide detailed information on how it was formed, the amount of data, and whether similar datasets (open source or proprietary) exist in the literature.
· For the training and testing of the neural network, the authors should clearly define the type of input provided to the network, whether 2D or 3D. This aspect is not well-defined. Additionally, they should specify how many data points are used for training and testing. Are the testing images completely unknown to the network? It's crucial that the network can classify the presence of black ice in contexts not seen during training.
· A more extensive analysis and comparison with the state of the art should be conducted, possibly including a comparative table.
· Many figures require more detailed descriptions, such as figures 2, 3, 4, 16, and 17.
Secondary Comments:
The authors should significantly improve the quality of the paper's presentation. Within the paper, there are various issues:
· The authors should describe the paper's structure in the introduction.
· The paper lacks uniform margins.
· Acronyms should be defined in the text the first time they are used, and a table of acronyms could be included.
· References to figures and tables are inconsistent, with variations like "Fig.," "figure," and "Figure" used.
· Paragraphs should not be used to define the names of neural network layers (e.g., 4.1, 4.2, 4.3, etc.). It would be better to use Table I and define a single name for the paragraph, such as "Neural Network Architecture."
· The same applies to section 5.2 (Visualization of Feature Maps). It doesn't make sense to have a subsection for each feature map.
· The bibliography format should adhere to the journal's template, and the paper should align better with the suggested journal template overall.
· Many figures have poor quality:
o Figure 1: Define the labels (and units) of the Cartesian axes clearly.
o Figure 2: Include "(a)" and "(b)" in the two photos and provide corresponding descriptions.
o Figures 13 and 14: The text in the legend is too small. Additionally, Figure 14 is not visible.
o Figures 13, 14, and 15: The text labels for the x and y axes are too large relative to the graphs.
· In Tables 2, 3, and 4, the F1 score values are written in decimal format, while others are in percentage. To maintain uniformity, it should all be expressed as percentages.
· In the text, section 5.1, there are references to formulas using numbers, but there are no reference numbers in the formulas. These reference numbers should be added.
These suggestions and comments aim to help the authors improve the clarity, quality, and comprehensibility of their research paper.
Comments on the Quality of English Language
Moderate editing of English language is required
Author Response
Dear Reviewer,
Thank you so much for taking time amidst your busy schedule. I appreciate your comments and feedback on our manuscript. I have corrected your mentioned point and modified the manuscript accordingly.
Thank you again for your valuable input on our manuscript which helped us enhance the quality of the paper.
Sincerely,
Chaitali Bhattacharyya
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for AuthorsRemarks (P: Page number, and L: Line number in the manuscript)
· P2: Contribution part, 2, 3 & 4 cannot be considered as contribution.
· P3: no citation for the statement: “mmwave technology is hard to manage in rain 98 and fog when the black ice forms. Because it suffers from attenuation in these kinds of situation. Compared to hyperspectral image the data are also a bit complicated for pre-100 processing and other custom settings.”
· Nothing is mentioned in the text about Figures 2, 3, and 4.
· No background about what is meant by the phrase “data cube”?
· First paragraph of the “Methodology” is ambiguous. There is no clarification of the parameters used, e.g., X, P, S, and B.
· L217: “Total parameters are 6,629,818.”
o Is this right? More than 6m parameters!!
· P8: Numbering Figure 9 and Figure 8 is incorrect.
· L238: “To understand a material, a good way is to see how the light works on the material.”
o Unclear statement.
· L265: “the equation 3, 4 and 5,”
o There are no equations with these numbers.
· Figure 14 is cropped.
· L276: “Table - II-IV shows”
o Incorrect table numbering.
· Figures are appearing before referencing them in the text.
· In the text, Tables are referenced using Roman numerals.
· L331: “Figure 15 shows the features read onto the black ice portion of the image.”
o Incorrect figure number.
· Generally, there are flaws in the content and organization of the paper.
Comments on the Quality of English LanguageGrammer and Spelling Errors
· “is not only cause damage to vehicles passing over that spot, but it puts lives at risks.”
· Considering the survey result from Korea Transport Institute [Figure-1], in the initial phase, the main aim of the paper was to collect data at daytime.
· Considering the survey result from Korea Transport Institute [Figure-1], 71 in the initial phase the main aim of the paper was to collect data at daytime.
· P3: In Fig. 1, in Korea yearly traffic accidents caused by black ice for 24 hours are shown [19].
· P3: From the figure 1 given by the official source
· P3: Because it suffers from attenuation in these kinds of situations.
· L137: information. which will also
· L189: “the output is reshaped to compress the final two 2D con-189 volution operations are connected by this.”
o Incomplete sentence
· L220: “used EarlyStopping o halt the training”
· L223: “Table 1. shows the summary of the whole model for black ice”
o Table title
· L250: “figures 12, 13, and 14”
· L265: “the equation 3, 4 and 5,”
· L298: “table -II - IV”
· L324: “in figure 16.”
· L377: “which means that it has string”
· Generally, the paper must be subjected to further revising and proofreading in terms of English grammar
Author Response
Dear Reviewer,
Thank you so much for taking time amidst your busy schedule and give detailed feedbacks on our manuscript. I appreciate your comments and concerns and corrected your mentioned points and modified the manuscript accordingly.
Thank you again for your valuable input on our manuscript which helped us enhance the quality of the paper. If there is anything else we should modify please let us know.
Sincerely,
Chaitali Bhattacharyya
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsThe authors addressed all of the questions clearly. The revised manuscript is appropriate for Applied Sciences.
Author Response
Dear Reviewer,
We are grateful to hear that you found our revised manuscript to be appropriate for publication.
Your insightful feedback helped us improve the quality and clarity of our manuscript.
Thank you again for your valuable time.
Sincerely,
Chaitali Bhattacharyya
Reviewer 4 Report
Comments and Suggestions for AuthorsIncorrect figure number!!
For black ice on the box, the same performance measurements were also 0.93. After the whole process, predicted images are in figure 15 for 10 meters, 20 meters, and 30 meters, respectively.
Comments on the Quality of English Language
The paper still have some flaws in terms of the grammar and writing style. For example,
- Tables 2,3,4, should be written as Tables 2, 3, and 4.
- A sentence is started as: figures 13,14,15 depict ...
- From the figure 1 given by the official source we can see 80-85% of
the accidents due to black ice in 24 hours happen in daylight In [20], multi-wavelength,
Kindly, the paper should be revised thoroughly.
Author Response
Dear Reviewer,
Thank you so much for dedicating your valuable time for our manuscript and providing your insights and detailed feedback. This has helped us to enhance the paper organization and writing.
We have taken every point into careful consideration and modified it accordingly.
Thank you again for your time.
Sincerely,
Chaitali Bhattacharyya
Author Response File: Author Response.pdf
Round 3
Reviewer 4 Report
Comments and Suggestions for AuthorsKindly, there are many issues that are not yet clarified.
Some (not all) of the notices:
Figure 4 is reffered in the text before Figure 3, and Figure 2.
The images were taken at (distances of) 10 meters, 20 meters, and 30 meters
yellow (color) represents the concentrated black ice and green (color) represents
Repetitions:
The data was labeled by MATLAB as shown in Figure 3. In Figure 3, the ground truth image shows black ice sample on the road and concentrated black ice sample on the box: yellow represents the concentrated black ice and green represents black ice spread on the road. Figure 3 (which part A or B) illustrates two black ice samples: one is on the asphalt and the other one is inside the box.
Unclear repetatition of the statement:
The matrix (which matrix?) captures the covariance between the bands
To implement the model, the data cube was divided into 3D overlapping patches. During classification, the hyperspectral image data cube was divided into small 3D patches.
the data cube (what is meant by data cube?) was divided into 3D overlapping patches.
What are the parameters in: S×S×B ?
...
Comments on the Quality of English Language
Some (not all) of the notices:
In Figure 1, yearly traffic accidents in Korea caused by black ice for 24 hours are shown [19]. According to the graph, majority of (missing: the) accidents caused by black ice occur in the 72 morning from 6 AM to 10 AM. From Figure 1, which is provided by the (an) official source
First(,) we have trained
Using a millimeter wave sensor with a deep learning model (a 1D Convolutional Neural Network)(,) another detection method achieved 98.2% accuracy 80 [23].
Compared to hyperspectral images(,) the data are also a bit complicated for preprocessing and other custom settings.
Figure 2. A: The camera set up; B: The black ice set-up at 10 meters distance from the camera.
process.Hyperspectral image data
We aim to detect both types (samples)
After these (this) steps, we calculated the eigenvector
to extract the bands, containing the majority
it (It) helps to reduce
n th (nth)feature map of m th (mth)
Overall, The input data is initialised
In summary, some feature(s) does have mean values
low mean values for example feature map 7
...
Author Response
Dear Reviewer,
Thank you so much for dedicating your valuable time to our manuscript and providing your insights and detailed feedback again. We have carefully read your comments and modified the whole paper.
We have attached the document. In the document, we have highlighted the modifications.
Thank you again for your time and consideration
Sincerely,
Chaitali Bhattacharyya
Author Response File: Author Response.pdf
Round 4
Reviewer 4 Report
Comments and Suggestions for AuthorsNotice this is the fourth time to review the manuscript, and it still has some grammar and writing issues.
Comments on the Quality of English LanguageSome examples are listed below:
1. As stated in the Section 6,
2. The macro average and weighted average for precision as well as the F1-score is 0.97.
3. Figure 16. The predicted ground truth of black ice: (a) at 10 meters (white is concentrated ice on box and gray is the ice spread on the road), (b) 20 meters and (c) 30 meters (White is ice spread on the road and the gray is the concentrated ice on the box.
4. Feature map 8 moderate magnitudes and moderate variability around the mean.
5. With low mean value, feature map 4 indicates that the activations are comparatively small on average and std value suggests that the activations are concentrated around mean.
6. Feature map 10 and 11 have a moderate spread and variability around the mean.
7. Observing the value (which value) in Figure 17, it might indicate 304 that this feature map is capturing binary or threshold patterns.
Author Response
Dear Reviewer,
Thank you so much for dedicating your valuable time to our manuscript and providing your insights and detailed feedback again. We have carefully read your comments and we found that we had few grammar issues in results and visualization sections. We have fixed them according to your comments.
We have attached the document. In the document, we have highlighted the modifications.
Thank you again for your time and consideration
Sincerely,
Chaitali Bhattacharyya
Author Response File: Author Response.pdf
