PolSAR Image Classification by Introducing POA and HA Variances
Round 1
Reviewer 1 Report
General Comments:
Overall, this paper adds nicely (if modestly) to the body of knowledge in polarimetric SAR image classification. The significance of the paper could be elevated if the authors took an additional step as discussed in section 5.3 to apply texture features to the backscatter power to better discriminate between areas designated as 'forest' and 'oriented buildings'.
Specific Comments:
Page 1, line 44: Convention when referencing a paper by two authors is to cite both authors, e.g. "Freeman and Durden" instead of "Freeman et al".
Page 4, line 142: Section 2.2: Your work needs to be easily reproducible by others. Please show mathematical formulae for rotations.
Page 6, Table 1: It would be helpful to the reader to add incidence angle, image dimensions and orientation (e.g. NW to SE) to this table.
Page 7: Figure 3(b) is actually an image of the San Francisco Bay area in California, and the highlighted area of study is the city of Oakland (not Richmond).
Page 7: Figure 4. The proper designation for the vegetation covered hills that surround the San Fernando Valley is 'Chapparal'. Chapparral is a type of low scrub forest, consisting of dense shrubs and dwarf oak trees common to the American SW. See e.g. https://www.researchgate.net/publication/45639859_Nitrogen_critical_loads_and_management_alternatives_for_N-impacted_ecosystems_in_California/figures?lo=1
Page 9, Figures 6 and 7: Histograms showing the POA and HA variance for different terrain types in this image would be helpful for interpretation of these results. Fig. 14 contains histograms but only of the POA variability.
Page 10, Table 4: Table 4: These classification results are good, and a significant improvement over results using just the results from the Freeman-Durden decomposition. However, there is still quite a lot of confusion/mis-classification of the oriented buildings as forested areas.
The paper could be improved (and have greater impact) if the authors could show results using other texture measures to further improve the classification results. Gridded streets with oriented buildings appear quite different texturally from naturally occurring vegetation. Texture measures based on the total power in the higher resolution image data might help. See e.g. https://www.researchgate.net/publication/224127130_SAR_Image_classification_using_textural_modeling
Page 11: Figure 8: This image, which is really of Oakland, CA, appears to be flipped about the vertical axis. The designation of the nearby hills as 'forest' is technically correct, though to be precise the land cover type is more completely described as 'mixed hardwood forest'. See again https://www.researchgate.net/publication/45639859_Nitrogen_critical_loads_and_management_alternatives_for_N-impacted_ecosystems_in_California/figures?lo=1
Page 13: Figure 12. I don't know this site but would guess that the forested terrain is 'seasonal tropical forest' if it is naturally occurring. See e.g. https://www.researchgate.net/publication/225521221_Modeling_distribution_changes_of_vegetation_in_China_under_future_climate_change/figures?lo=1
Page 14: Figure 14: HA histograms are not shown. Not clear which image these histograms are extracted from.
Page 15, section 5.3: Section 5.3: Missing references. A lot of work has been done on using texture in SAR images to improve classification. E.g. https://www.researchgate.net/publication/224562741_Modeling_and_Simulation_of_SAR_Image_Texture
Page 16, section 5.3: Section 5.3: "In the future, we will make full use of orientation and scattering power texture 422 information to further improve the land cover classification accuracy. ". Why not do this now? It has the potential to increase the impact of your paper significantly.
Your use of English is quite good, but could use a little help from a good editor.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The equations for calculating POA and HA should be provided.
Whether it is necessary to classify the buildings into orthogonal and oriented. And what about buildings with other oriental angles?
The calculation of POA variance looks strange, for instance, -45 and 45 degree is 1 and 10, holding the largest label difference, however, their angle distance is really close.
The HA is not shown in Figure 14, moreover, the variance of POA and HA should be analyzed
Besides, freeman decomposition parameters, other polarimetric features for urban classification on classification should also be considered, such as H/A/alpha, DoP, polarization ratio, etc.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 3 Report
See the attachment file.
Comments for author File: Comments.pdf
Extensive editing of English language required.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Thanks for your response to my comments on your earlier draft. I think the addition of the texture features in the final classification significantly improves the paper.
There are still a couple of typos in your MS, but I assume the editorial staff at Remote Sensing will help with those. I look forward to seeing the final version in print.
Author Response
Point 1:Thanks for your response to my comments on your earlier draft. I think the addition of the texture features in the final classification significantly improves the paper.
There are still a couple of typos in your MS, but I assume the editorial staff at Remote Sensing will help with those. I look forward to seeing the final version in print.
Response 1: Thank you for your comment. Thank you for your support of our work.
Reviewer 2 Report
The reviewer is generally satisfied with the revised version.
It is suggested to improve the English expressions, to avoid reducency and non-native expressions.
Author Response
Thank you for your comment. Thank you for your support of our work.
This manuscript has been polished by MDPI English Editing Company. We have checked the manuscript carefully.
Reviewer 3 Report
1. In Introduction: some relevant refs regarding cross scattering model and rotated dihedral scattering model should be cited, such as 10.1109/LGRS.2015.2487450, 10.1109/TGRS.2023.3257773, and 10.3390/rs15010101.
2. For my previous comment #9, I don’t think that the authors have responded my query as they only mention that the introduction of more features are helpful for classification. This is just natural! The proposed method has not been compared with the latest methods and the involved comparison is rather simple and primary. This is because the incorporation of extra information can always lead to better results, which cannot adequately demonstrate the advantages. In addition, although the adopted decomposition is An’s reflection symmetric approximation method, the essence is still the three-component decomposition, which does not consider the scattering behavior induced by oriented buildings.
3. For my previous comment #11, the authors have failed to elaborate the novelty of their work. As mentioned, it only proposes the variance parameter of POA and HA, while the designed division pattern is also similar to the POA randomness. The authors argue that texture information can be used as a supplement to scattering power in PolSAR image classification. I don’t think this is a tellable innovation as it is well known. Although the authors have conducted sufficient experiments with three adopted PolSAR datasets, the comparison is less convincible to demonstrate the superiority, as mentioned above.
Minor editing of English language required
Author Response
Please see the attachment.
Author Response File: Author Response.docx
