2chADCNN: A Template Matching Network for Season-Changing UAV Aerial Images and Satellite Imagery
Round 1
Reviewer 1 Report
I really liked the work in general; however, I would have liked to know the behavior of the proposed method on exclusive crop surfaces such as vineyards and forest areas. Has the test been conducted on these types of covers?
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
In this paper, authors proposed an algorithm for UAV aerial and satellite images in different seasons (spring, summer, fall and winter). The paper is well organized and well written. Authors proposed to use a 2 channel CNN with attetion blocks in order to extract local and glogal features and then using a similarity metric, authors determine probability of image matching.
The paper lacks of novelty, authors used an achitecture of CNN of 6 convolutional and two dense layers.
According to the paper, authors propose a 2chADCNN, however, it is not clear how this two channel CNN perform the image matching task.
The figure 2 is confusing, I suggest to redesign it to have a better understanding
In the manuscript, authors say lower layers extract local features and higher layers extract global features, how do you know that? Furtermore, in section 3.1, author say they proposed a scaling factor. How did you come out with this proposal?
I suggest to write more about GSD
In the paper I did not find information about the reason of the proposed network achitecture.
In the paper is proposed the use of channel attention and spatial attention modules, however, in the experiments authors did not show how these modules help to improve the image matching task. I suggest to include in the results section, experiments of the attention modules.
I suggest to include in the conclusion section the reason why, by using the proposed method, the image tasking is improved
Minor corrections are necessary
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Review report for drones-2537062
Manuscript ID: drones-2537062
Type of manuscript: Article
Title: 2chADCNN: A template matching network for season-changing UAV aerial images and satellite imagery
Authors: Yan Ren *, Yuwei Liu, Zhenjia Huang, Wanquan Liu, Weina Wang
1. I’m happy with Section 1: Introduction.
· The author made a clear and intuitive introduction, which well-covered the background, motivation, research gap, scope, technological contributions, and manuscript layout.
· To improve the introduction, the only suggestion I can make is: the author should make a clear and firm definition of “image matching”.
2. Section 2: Related Work
· Paragraph 1 does not sound very making sense to me. The author summarized image-matching methods as “tradition techniques” and “CNN-based approaches”. Then, only “tradition techniques” was described in the rest of Paragraph 1. Then, Paragraph 1 jumped to some applications.
Paragraph 1 described three meanings in a very short paragraph. The logic sounds very confusing.
· The author divides grayscale-based image matching from feature-based matching (Paragraph 3) and traditional matching (Paragraph 4). I’m not fully convinced by the way of the author’s classification. I will suggest the author provide clear definitions for each class to help the reader understand what’s the reason for the author’s classification.
§ SIFT and ORB also apply grayscale, which makes them different than grayscale-based image matching?
§ The 1st sentence of Paragraph 3 requires a reference to support.
§ The author applied too much “tradition” to indicate the category of image-matching technology, which cause confusion.
· Section 2 provided a very detailed literature survey, which can be further improved by addressing some minor corrections.
3. Section 3: Proposed Methodology
· Figure 1: The author should use sub-titles (like (a), (b), and (c)) under the figure to indicate sub-figures. The “left”, “middle”, and “right” is not clear way to describe the figure.
· Same as in the rest of the Figures.
· I’m happy with the rest of the methodology descriptions.
4. Section 5:
Table 1 and Table 2 show a huge improvement from the proposed method. It might come from the author did not apply the latest method, and I wonder what’s the reason for it. I checked the comparison methods the author applied, Ref. [13] is the latest, while the performance seems very poor. Ref. [35] and Ref. [36] are the methods within 3 years, which achieve relatively good performance.
It’s okay if the author only uses these comparisons. But, to further improve the quality of the research, the author may implement some more recent methods (better choose from prestigious journals or conferences), and do the comparison to the proposed method to better justify the performance of the proposed method.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
The authors now presented an improve version of the paper, correcting all the observations made
Good use of english. Minor corrections needed
