Next Article in Journal
Assessment of Estimated Phycocyanin and Chlorophyll-a Concentration from PRISMA and OLCI in Brazilian Inland Waters: A Comparison between Semi-Analytical and Machine Learning Algorithms
Next Article in Special Issue
A Clutter Parameter Estimation Method Based on Origin Moment Derivation
Previous Article in Journal
Possible Future Climate Change Impacts on the Meteorological and Hydrological Drought Characteristics in the Jinghe River Basin, China
Previous Article in Special Issue
A Modified 2-D Notch Filter Based on Image Segmentation for RFI Mitigation in Synthetic Aperture Radar
 
 
Technical Note
Peer-Review Record

Recognizing the Shape and Size of Tundra Lakes in Synthetic Aperture Radar (SAR) Images Using Deep Learning Segmentation

Remote Sens. 2023, 15(5), 1298; https://doi.org/10.3390/rs15051298
by Denis Demchev 1, Ivan Sudakow 2,*, Alexander Khodos 3, Irina Abramova 4, Dmitry Lyakhov 5 and Dominik Michels 5
Reviewer 1:
Reviewer 2: Anonymous
Reviewer 3:
Remote Sens. 2023, 15(5), 1298; https://doi.org/10.3390/rs15051298
Submission received: 15 January 2023 / Revised: 18 February 2023 / Accepted: 24 February 2023 / Published: 26 February 2023
(This article belongs to the Special Issue SAR-Based Signal Processing and Target Recognition)

Round 1

Reviewer 1 Report

This manuscript proposes a two-stage deep learning-based recognition framework for automated tundra lake recognition by using Sentinel-1 SAR data. This application work is interesting and looks effective. I have the following concerns with the manuscript:

1)     Besides U-Net, there are many other deep learning frameworks. Please explain the reason why U-Net is chosen. It is better to compare some other deep learning method and other classical image segmentation methods.

2)     In my opinion, using only HH or HV channel image for tundra lake recognition is also effective. Please state the reason why incidence angle information is introduced for the proposed method.

Author Response

please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

The whole article is full of practical engineering significance, while the following aspects need to be improved:

 1)     The author claims that at present, it mainly depends on manual analysis, but as far as I know, there are also some objects that are extracted and recognized using machine learning. It is suggested to add relevant references in the introduction.

2)     In section 2.3 and 2.4, the manuscript explains less about the innovation of the proposed algorithm. It is suggested to describe theory or method of these two key steps. For example, what semantic information is used in semantic segmentation?

3)     In Fig.2(D), why combine HV, HH and Ia into a 3-channel image? What is its physical meaning? What benefits does it have for tundra lakes shape and size recognition? In addition, there is a writing error in the title of Fig.2(D).

Author Response

please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

The paper presents the application of U-net segmentation, followed by watershed semantic segmentation, for retrieving the shape of tundra lake shapes, based on SAR images.

The paper is in general well written and it deserves, in my opinion, to be published in Remote Sensing.

My comments are the followings:

-        All the methods and indexes used in the paper are briefly described except the U-net network. I think that a brief description of the idea behind Un-net networks and of their architecture would be beneficial for the reader.

-        There is a relatively large discrepancy between the values of the Jaccard index for training and validation data sets (Fig. 5) – could you comment on this ?

-        In the Discussion section there is a quite long discussion on the importance of the fractal dimension in characterizing the evolution of the complexity of the lakes shape. However, the data presented in the paper is nto really relevant for this discussion. I would recommend to shorten this discussion.

-        I do not understand what the authors want to say in the following paragraph: “It seems that high-resolution images and sophisticated image analysis methods are not appropriate to catch the power law features of tundra lakes”. Why a high resolution image and a set of sophisticate methods would not help to retrieve any kind of information ?

Author Response

please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The manuscript has been sufficiently improved and can be accepted in present form.
Back to TopTop