PolSAR Image Building Extraction with G0 Statistical Texture Using Convolutional Neural Network and Superpixel
Round 1
Reviewer 1 Report
The paper proposes a building extraction approach from PolSAR images with statistical texture and polarization features to train a CNN model.
The training test positive and negative samples selection needs clarification:
327 - For the selection of negative samples in the non-building area, a certain range is defined. The positive samples with the same number are randomly selected.
It´s not clear what is the percentage of non-building samples selected. Also 15%?
It´s not clear also if the CNN training samples are excluded for the calculation of AR, FAR and F1.
Some statements are supported only by one reference. It would be interesting if there were 2 or 3.
Some sentences need grammatical improvement:
67- and the null angles of and
117- of the target can represent by a 2 × 2 complex matrix
260 - After the After the calculation, update the center of each superpixel. Next, repeat the
284 - the proposed building extraction method is compared with the existing building extraction method.
291 - Three different study areas and data sets are used applied in this paper. The optical
It´s ground or range? 305 - and the azimuth resolution and the ground resolution are both about 8 meters
Author Response
Thank you for your suggestions. Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
This paper concerns building extraction from PolSAR satellite imagery. The authors make use of a novel deep-learning method together with super-resolution. I would not say that the presented method showed strong originality, but the results are convincing and technical description and experimental validation are very strong. The authors present the technique in a clear way that strongly justifies the approach and evaluate the work on three datasets of relevance to the problem with comparison to relevant prior work.
The discussion section of the document is very well-done with pseudo-ablative studies and a detailed insight into the shortcomings and strengths of the algorithm. Overall I feel that this work is of a high quality and suitable for publication. One minor correction that I suggest is:
Page 8, lines 263-263: "Combine the number of pixels with less than a certain
number of superpixels into the nearest superpixel to obtain the final PolSAR superpixel image." seems to be an error. Shouldn't it be:
"Combine the number of superpixels with less than a certain
number of pixels into the nearest superpixel to obtain the final PolSAR superpixel image."
In addition, there are minor grammatical issues in places. I suggest that the authors do a thorough proof-reading.
Author Response
Thank you for your suggestions. Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Please see my comments on the attached file
Comments for author File: Comments.pdf
Author Response
Thank you for your suggestion. Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
It is an interesting research by itself. Though there is no much novelty in terms of methods, it is a great application and outcome for the great experimental results from the PolSAR data. I would suggest the acceptance of the manuscript for the publication. (No file attached)