Next Article in Journal
Synthesis of Leaf-on and Leaf-off Unmanned Aerial Vehicle (UAV) Stereo Imagery for the Inventory of Aboveground Biomass of Deciduous Forests
Next Article in Special Issue
Integrating Multitemporal Sentinel-1/2 Data for Coastal Land Cover Classification Using a Multibranch Convolutional Neural Network: A Case of the Yellow River Delta
Previous Article in Journal
Mapping Paddy Rice Using Sentinel-1 SAR Time Series in Camargue, France
Previous Article in Special Issue
Mapping Winter Crops in China with Multi-Source Satellite Imagery and Phenology-Based Algorithm
Article
Peer-Review Record

Smallholder Crop Area Mapped with a Semantic Segmentation Deep Learning Method

Reviewer 1: Anonymous
Reviewer 2: Anonymous
Reviewer 3: Moshiur Rahman
Received: 19 February 2019 / Revised: 2 April 2019 / Accepted: 3 April 2019 / Published: 11 April 2019

Round 1

Reviewer 1 Report


I consider the research as being a very interesting and well documented paper. As such, I only have few suggestions for authors:

In Abstract, revise the 1st paragraph as something is wrong, especially the last statement (is a powerful tool towards this goal) does not link with the begining part.

Legend of Figure 6: Change Ture to True and nagative to negative.

lines:

 - 284 word segmantation should be changed to segmentation

 - 293 word aumatically should be changed to automatically AND word resluts to results

In title of tables 5 and 6 change meachine to machine.

In the study area section, I consider the authors should explain why Baodi was chosen as an area that is representative for China (besides being a food and cotton producer). Are many regions similar so that their research can be extended to others?

Also, in Data section maybe is useful to present a statistic about how many smallholder family farming exist in China and/or Asia. This way the audience can understand why is this research representative for China or other countries.

In the conclusions, the authors should explain how can their methodology be applied to other similar areas.

Author Response

Thank you for the comments which are highly insightful and enabled us to improve the quality of our manuscript. The point-by-point response please see the attached file.

Author Response File: Author Response.pdf

Reviewer 2 Report

Smallholder crop area mapped with semantic segmentation deep learning method

 

The research focused on validating deep learning as a tool to extract and classify crop area in high resolution satellite imagery.  It specifically focused on red, green, and blue band imagery and small fields, located in China.  The deep learning technique performed better than commonly used machine learning tools such as maximum likelihood, support vector machines, and random forest.  Generally, appropriate methods were used to obtain the results.  The document contains excellent figures. 

 

Major issues with the manuscripts are as follows: (1) it lacks necessary details in several sections (see comments below related to each section) and (2) it contains typos and grammatical errors that affect the overall content of the document.

 

Abstract

1.      Insert a sentence after the objective that indicates RGB imagery will be evaluated in the study.  Better yet, indicate satellite imagery was evaluated in the study.  That is important to the reader of the manuscript. 

2.      The Abstract needs more detail, for example state specifically which traditional methods were compared to the DeepLabv3+. 

3.      OA = overall accuracy, spell out in Abstract

 

Keywords

1.      Crop area and deep learning are used in the title. Consider replacing those key words. 

 

Introduction

1.      Pg. 2 of 18, lines 46-63:  Paragraph is too long.  Divide it into two paragraphs.

2.      Pg. 2 of 18, lines 70-82; Pg. 3 of 18, lines 83-90:  Paragraph is too long.  Divide it into two paragraphs.

 

Study Area and Data

1.      Pg. 4 of 18, line 129:  1 m resolution or was it close to 1.85 m resolution. Was the multispectral imagery fused with the panchromatic image? If yes, then insert that information into the manuscript.  Overall, what satellite product did you receive from the vendor.    

 

Method

1.      Pg. 4 of 18, line 135:  Reference should be for Figure 2 not Figure 1. 

2.      Pg. 5 of 18, lines 153-154:  Insert a reference related to PASCAL VOC 2012 and Cityscapes datasets.

3.      Pg. 6 of 18, lines 172-173:  Insert reference related to the Softmax function. 

4.      Pg. 6 of 18, lines 177-178:  Spell out meaning of GPU; Graphic Processing Unit

5.      Pg. 9 of 18, lines 222-223:  Add 1 reference for each classification procedure or 1 book would suffice if it included a general description of each classification procedure.

 

Results and Discussion

1.      Results are provided but there is no discussion or comparison to other research studies. 

2.      Discuss the results.     

 

Tables

Overall, tables should stand on their own without the reader referring to the text to find the meanings of abbreviations.

1.      Table 1. In the table header, spell out meaning of abbreviations and then insert abbreviations in parenthesis…. Maximum likelihood (ML), ……. Insert meaning of RBF as footnote.   

2.      Table 2.  Insert footnote referencing the meaning of CA, Non_CA, and OA.

3.      Table 3.  Insert footnote referencing the meaning of U-Net, PspNet, SegNet, DL, AVG, and OA. 

4.      Table 4.  Insert footnote referencing the meaning of U-Net, PspNet, SegNet, DL, AVG, CA, and Non_CA. 

5.      Table 5. Insert meaning of CNN in the figure caption. Insert footnote referencing the meaning of Val, OA, AVG, ML, SVM, and RF.

6.      Table 6. See comments for Table 5, also include CA and Non_CA meanings in the footnote.     

 

Figures

Overall, figures should stand on their own without the reader referring to the text to find the meanings of abbreviations.      

1.      Figure 2. Spell out CA in the figure caption. 

2.      Figure 3. In figure caption add meaning for Conv, example Conv = spell out the word.

3.      Figure 4. In figure caption add meaning for CA, RGB, and GT, CA = spell out the word or spell out word and put meaning in parenthesis, crop area (CA), red, green, blue (RGB), and ground truth (GT)

4.      Figure 5. In the figure caption, add the method used to derive the results.

5.      Figure 6. Correct spelling, Ture should be True, nagative should be negative. In figure caption add meaning for UNet, PspNet, SegNet, DLv2…v3+

6.      Figure 7. See comments for Figure 6. Also, spell out ML, SVM, and RF in the figure caption.      

 

References

1.      Pg. 17 of 18, line 371:  Correct formatting of authors names. 

2.      Pg. 17 of 18, lines 387, 405, 428-429:  Correct formatting of journal title.   

 

General comment

1.      Changed researches to studies

2.      Spell out abbreviations the first time mentioned in text


Author Response

Thank you for the comments which are highly insightful and enabled us to improve the quality of our manuscript. Please see the point-by-point response in the attached file.

Author Response File: Author Response.pdf

Reviewer 3 Report

Please see the attached file


Comments for author File: Comments.pdf

Author Response

Thank you for the comments which are highly insightful and enabled us to improve the quality of our manuscript. The point-by-point response please see the attached file.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Thanks for the amendments according to suggestions.

Author Response

p.p1 {margin: 0.0px 0.0px 0.0px 0.0px; font: 11.0px 'Times New Roman'; color: #000000; -webkit-text-stroke: #000000} span.s1 {font-kerning: none}

Thank you for the helpful comments to improve the quality of our manuscript.

Back to TopTop