Detecting Moving Trucks on Roads Using Sentinel-2 Data

Round 1

Reviewer 1 Report

This manuscript addresses moving truck detection with Sentinel-2 Data showing overall performance 82%. This study has its own novelty analzying moving trucks for object detection. I think this manuscript can be presented in the current form. One minor comments - the brief numerical results can be included in Abstract.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

This research presented a method for detecting moving trucking on roads using Sentinel-2 satellite imagery. This proposed methods explores a temporal sensing offset of the Sentinel-2 multispectral instrument, causing spatially and spectrally distorted signatures of moving objects. This proposed method integrates a random forest classifier trained on natural color and near-infrared spectra of 2,500 globally labeled targets with a spatial object extraction method. The method was validated with 350 globally labeled targets, presenting a result of mean F1 score of 0.74. The reviewer believes that the current version of the manuscript is not yet ready for publication; the authors are encouraged to consider the following comments and suggestions and revise the manuscript accordingly.

1. The authors should consider streamlining the Abstract section. Currently, the Abstract section is not in a natural flow and it provides limited information. The Abstract section should be focused on providing a summary of the study and discussing the contribution to the body of knowledge.
2. The authors should also consider splitting the Introduction section into two sections, including an Introduction section and a Background (or Related Work) section. The introduction section should focus on introducing the research objectives and research questions, while the Background section should focus on reviewing of related work and defining the research gaps. The authors should also review more related literature.
3. The authors need to provide more information to explain why a trucks can be detected in a lower spatial resolution image (20m). The authors should also read and cite the paper of “on the nature of models in remote sensing.” The authors are using an L-model.
4. The authors need to provide a supplemental document to show more detailed processes for the proposed random forest classifier. This document should show all the parameters and processes involved in the classification process.
5. What is feature stack? This is the first time I hear about this term. Do you mean models? The authors need to provide more explanation for terms that used in the manuscript.
6. The authors should have a professional editor proof-read their manuscript. There are many grammar errors in the current manuscript and it needs to be improved.
7. The authors should go through the equations to ensure all symbols have been appropriately denoted.
8. The authors should discuss the future of the proposed method. For example, recently drones of unmanned aircraft systems (UAS) have become increasingly popular as a data collection platform. The authors should discuss the potential of applying the proposed method on drone imagery. The authors should read and cite the paper of "the impact of small unmanned airborne platforms on passive optical remote sensing: a conceptual perspective."
9. Most of the figures need to be improved. For example, in Figure 1-4, the reviewer has to zoom in at least 150% to be able to read. If at all possible, please create vector images for readability. In addition, the authors need to improve the tables. Some tables are even not necessary.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

  The paper describes a method to count trucks on the road using images from Sentinel-2 satellite and a Random Forest classifier. The paper is well structured and the method seems coherent.

  Here are some comments to hopefully help improve the paper.

  1) The results obtained are not so good, and the performance varies as a function of the state of the road and other variables. Based on the results shown it is very optimist to affirm the method can be used for truck counting when an accurate counting is required.  Nonetheless, it still has merit as a first approach.

    1.1) I wonder if the authors considered using other classifiers for the same purpose, or a deep neural network for object recognition, at least as future work.
    
    1.2) It is not totally clear if the dataset was balanced, containing sufficient examples of all types of roads, weather conditions, etc.. The composition of the dataset could be explained in more detail.
    

  2) In Section 2.5.4 is affirmed the hyperparameters were optimised using random search. The search limits could be clarified. Also, if RS was chosen for any special reason, among the many options for optimising hyperparameters.

  
  3) Equations should be revised and clarified.
    - 2res in Equation 2 is confuse. 
    - In Equation 11, usually the quotient is multiplied by 2 to have F1 in the range [0, 1].

  4) The paper is in general well written, but there are still some parts which are confuse and typos along the paper. Data is a plural word, not singular, which is datum.
 
 
  5) The flowchart in Figure 4 could have start and end. Same for Figure 7, which is even more difficult to follow.

  6) In the PDF I accessed some figures are cropped on the right. Namely figures 4 and 13.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The authors have addressed all my comments. 

Reviewer 3 Report

The paper looks better, with errors corrected and more complete now. In general I am satisfied with the changes and authors' responses.