Multi-Input Attention Network for Dehazing of Remote Sensing Images
Round 1
Reviewer 1 Report
The authors proposed a new method for removing haze in satellite images in this article. The authors should share the test data set they created for other researchers who will check this method and compare. This article needs proofreading for some minor English corrections. Some examples;
Line 2: brings
Line 4: remote sensing hazy image? > remotely sensed hazy images
Line 10: two “different” resolutions
Line 67: landsat8
Line 96: “outstanding results” seem to be very pretentious here.
Line 178: Do we need table 1?
Figure 3: downsamplling – upsamplling (extra “l”)
Author Response
Please see the attachment
Author Response File: Author Response.docx
Reviewer 2 Report
This paper presents a deep neural network for remote sensing image dehazing, and the novelty lies in using multi-band images (nine bands) instead of the conventional RGB images. After scrutinizing the manuscript, the reviewer has some comments as follows.
1. The authors only utilized three equations throughout the paper. Notably, only one equation is adopted to explain the proposed network architecture. This lack of mathematical expressions lowers the paper’s readability and reproducibility. Therefore, it is highly advisable to describe the proposed network architecture more precisely and concisely using mathematical expressions.
2. On line 288, the reviewer deems that it would be better to replace “gradient disappearance” with “gradient vanishing.”
3. It is unclear whether the authors have utilized a validation set to check whether the proposed network was over- or under-fitted. Please provide more details about the training process.
4. It is worth noting that SSIM and PSNR do not correlate well with human subjective assessments. Hence, it is recommended to adopt other metrics (such as FSIMc [a] and TMQI [b]) for performance evaluation.
[a] L. Zhang, L. Zhang, X. Mou, and D. Zhang, “FSIM: A Feature Similarity Index for Image Quality Assessment,” IEEE Transactions on Image Processing, vol. 20, no. 8, pp. 2378–2386, Aug. 2011, 10.1109/TIP.2011.2109730.
[b] H. Yeganeh and Z.Wang, “Objective Quality Assessment of Tone-Mapped Images,” IEEE Transactions on Image Processing, vol. 22, no. 2, pp. 657–667, Feb. 2013, 10.1109/TIP.2012.2221725.
Also, there is another metric dedicated to image dehazing, HDE [c], that can also be used for assessment.
[c] D. Ngo, G.-D. Lee, and B. Kang, “Haziness Degree Evaluator: A Knowledge-Driven Approach for Haze Density Estimation,” Sensors, vol. 21, no. 11, p. 3896, Jun. 2021, doi: 10.3390/s21113896.
5. Benchmark models are designed for natural image dehazing, and the authors have mentioned on lines 196–198 that natural image datasets could not be applied to remote sensing image dehazing. Therefore, before evaluation, those models must be “re-trained” with remote sensing image datasets.
6. There are eight possible variants in the ablation study, but the authors have only presented three of them. Please add the other five to gain more insights into the proposed network.
Author Response
Please see the attachment
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
The authors have addressed all comments, and thus the reviewer supports the paper's publication.