Classification of Road Surfaces Based on CNN Architecture and Tire Acoustical Signals

Round 1

Reviewer 1 Report (New Reviewer)

Dear Authors,

The article presents an important issue but it has to be improved.

1. Nomenclature including list of abbreviations has to be added

2. In lines 203-204 you write about the surface roughness. how it was evaluated that both surfaces have the same roughness. which parameters were evaluated 

3. What wavelet transformation parameters were used for the analysis and why these?

4. statistical evaluation of studies should be presented with special emphasis on cwt spectrograms

5. the discussion should be improved, extended

Kind regards

Reviewer

 

 

 

 

 

 

 

Author Response

Please find attached files.

Author Response File: Author Response.docx

Reviewer 2 Report (New Reviewer)

Tyre/road interaction noise feature recognition is used to distinguish different road surface types, which plays an important role in supplementing the vision of automatic driving and reducing the amount of data analysis. The study is very interesting. Comments are as follows: 

1. It is recommended to give a definition or description of “road surface classification” in this paper. Considering the contents of the paper, it seems the identification of road conditions (with or without snow). According to your research objectives, what other conditions should be included, in addition to snow?

 2. In Figure 5. The details of microphone fixing cannot be seen. How to ensure that the test is not disturbed by wind, snow or vibration brought by tires? What are the requirements for the setting positions? 

3. Low frequency noise (such as below 200Hz) is generally difficult to test accurately. What method is used in this study to ensure the test accuracy of low frequency noise? 

4.In Fig. 8, how to make sure that the differences between curves are not caused by test errors, because the differences between them are not large.

5. In addition to the test road, whether the accuracy of CNN model is verified on other practical roads?

Author Response

Please find attached files.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report (New Reviewer)

Dear Authors,

please provide a more specific response to the second comment in the previous review

How the variability of the distribution of irregularities in the sample was evaluated. You can not say that because it was the same asphalt road, the irregularities are the same. from a metrological point of view, such a statement can prove a lack of knowledge of the subject. measuring one surface in several places can result in completely different results. how it was evaluated that both surfaces have the same roughness. which parameters were evaluated

Kind regards

Reviewer

Author Response

Please find attached file.

Author Response File: Author Response.docx

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

The paper presents a road classification method using sound where a 2D input image is constructed from sound recordings using wavelets. The overall idea is very interesting. However, there are several aspect that requires improvement.

The very low level discussion of CNN has be removed as CNN are well described in literature. However, the actual architecture used is not presented. What are the filter size and number, how many Conv layers are there, is the final layer linear or with activation.

Only four classes are used and the structure of the classes need careful consideration. Why asphalt and snow classes only. Why separate the tyre into separate classes, while this has advantages it also makes the system tyre dependent and not necessarily robust to different tyres. This needs some discussion.

 

I have several questions about the data collection. Was the same road used for all tests (ie.e same road for all asphalt and another for snow) , was it done on the same day.
How much snow? was it always fresh snow or compacted after different runs? How would water affect the result or different surface roughness?
Was a single recording 60m?
At what speed and was it constant over recording? the same speed for all test etc? The authors noted that speed is one of the most sensitive aspects but not discussed how this was used in the analysis.

How was the mapping done from wavelet to RGB done? was it a colour map applied to the magnitude? Was there a normalization applied within the image or global normalizations i.e. white defined as predefined value for all tests or maximum of image etc.. More detail is required here

Comments for author File: Comments.pdf

Author Response

Please find attached file.

Author Response File: Author Response.pdf

Reviewer 2 Report

Page 1, Abstract, rows 12 – 13: wrong statement.

Page 1, Abstract, general: it does not include the essence of the paper.

Page 2, Figure 1, eq.1: the reference(s) has to be mentioned.

Eq.1: it is incorrect written.

Page 2, rows 74, 75 (ref. eq.2): the reference has to be mentioned.

Page 2, rows 75 - 79: to verify and to mention the reference.

Page 2, row 76: the x-axis coordinate (being = 0) does not depend on . To correct.

Page 3, Figure 2, eq.3 and beyond (in fact, for all the “Background Theory” part): the reference(s) has to be mentioned.

Page 5, row 152: “driver-induced” is not a source for tire-pavement noise; to correct!

Page 5, rows 159 - 162: the technical expression is not specific to the tire-road interaction research domain; it has to be reviewed. Also, ”proportional to the rpm of the tire” is incorrect – ”rpm” is a measuring unit – to correct!

Page 5, rows 165 - 178: to review the text; it is better to cumulate the previous paragraph with this one.

The part ”2.3 Tire–Pavement Interaction Noise” – what is its relevance versus the paper content?

# ”3. Experiment”: no information about absolute and comparative test speeds is presented; to provide information.

Page 5, row 186: why it was chosen the sampling frequency of 51200? To explain the reasons.

Page 5, # 3.1: the microphones were placed too close to the tires; to explain the reasons taking into account the acoustic field.

Figure 5: ”Experimental setup of instrument ………….” – to reformulate.

Page 6, #3.2: the paper has to specify details about the condition of the snow, as well as about the type of tires - at least if they are summer or winter model.

Page 6, row 192 (and Figure 6): ”snow” is not a test road – to correct.

Page 6, row 204: to replace ”rows” with ”columns”.

Page 6, row 205: ”The unit of the x-axis is second, and the unit of the y-axis is Pascale.” – to remove the text.

Page 6, #3.3: the data acquisition cannot be performed under 20 Hz (at least!) – depending on the microphone frequency response; also, the tire-road noise frequency spectrum usually is shown using A-weighted sound level – the paper presents linear sound level, thus the curves in Figure 8 are totally unusual.

Page 6, rows 209 – 216: the deductions presented are the result of inaccuracies in the measurement, so they are partially incorrect. Also, the explanation of the phenomena is simplistic and unfounded.

Page 7, Figure 8: the frequency scales have to be logarithmic.

Page 7, row 233: like above, it is not correct to investigate sound with frequency of 10 Hz.

Figure 10: the images have no axes.

# 6. Discussion - most of it is a summary of the paper.

There are no conclusions.

The purpose of the special analysis is not presented.

 

Author Response

Please find ttached file.

Author Response File: Author Response.pdf

Reviewer 3 Report

Dear Authros, the presented topic is very intersting and actual research field. However, there is lack of any presentation of the applicability of the obtained results. As it was stated at the beginning  of the paper, the classification of the road will be usefull for all the sensors in the car. I believe this classification will be of high interest of the NVH engineers also. Unfortunately, in the end only the confusion charts are presented. The paper requires a phisical interpretation of obtained data. How, the data are usefull for the sensors controll, NVH etc... What phisical characteristics can be extrated by this method?

The discussion/conclusion must be improved.

Author Response

Please fins attached file.

Author Response File: Author Response.pdf

Reviewer 4 Report

Dear authors,

the article is interesting. But I have some comments:

-  you mentioned 2 types of used tires, but you did not define the shape of the tire tread, the tread depth, how old the tire is, whether it is new, short-used or worn

- you mention the type of car, but you did not define the type of engine or the age of the car

- - in general, parameters / data about the road, plane, slope of the road, roughness of asphalt are missing, only the length is defined

-Fig. 9 and Fig. 10 are interesting, but are asked to be described in more details

 

Author Response

Please find attached file.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors have made slight changes to the script which answer most of the superficial questions but does not address the more complex questions.

The authors state that it the purpose is to classify snow roads, then still why split different tyres into different classes this does not make sense. Does mean you will have hundreds of classes for each possible tyre. If it is done for robustness then there are better ways. This requires a detailed discussion

 

In terms of the vehicle speed, the authors answer the basic questions but do not delve deeper in terms of the result of taking speed at one speed. Would this system work at 35 or 45km/h no disccussion surrounding this has been provided. The authors made it clear speed matters but take this no further.

Producing the colourmap is still not detailed the authors reply is "The scale of color map generated by CWT is sound pressure level. It is normalized
for 0 to 1"  How is the CWT used to create and RGB image. i.e. going from single pressure value to a tripple RGB value. Secondly normalised 0 to 1 using what? The max of the image or the maximum obtained over all images or something else.

 

 

 

Author Response

Please find the attached file.

Author Response File: Author Response.pdf

Reviewer 2 Report

Some (minor) points have been corrected. Also, few explanations tentative were presented. But the important suggestions have remained unsolved.

Author Response

Please find attached file.

Author Response File: Author Response.pdf

Round 3

Reviewer 2 Report

“pa” unit – To correct: “Pa” unit – pay attention to these “small” details in all paper text (page 8, Figure 9 etc.).

Conclusions have to be re-writing.